CN101561395B - Phytoplankton composition quick determination method - Google Patents

Abstract

The invention discloses a phytoplankton composition quick determination method, including the following steps of: (1) obtaining an original library of spectrogram of a fluorescence spectrum of the phytoplankton; (2) constructing higher-dimension orthogonal space, projecting the original library of spectrogram of the fluorescence spectrum of the phytoplankton to the high-dimension space, extractingthe fluorescence spectral feature of the phytoplankton and establishing a library of characteristic spectrum of the fluorescence spectrum of the phytoplankton; and (3) conducting discriminatory analy sis on the fluorescence spectrum of the phytoplankton sample and giving out the varieties and/or quantities of the phytoplankton contained in the measured sample. According to the composition characteristic of inshore phytoplankton of China and the structural characteristic of phytoplankton when red tide occurs, the method establishes a fluorescence identification method, which can determine the composition of phytoplankton at the class level in the situation that red tide does not occur and can identify and determine the phytoplankton to which red tide occurs at a genus level when red tide occurs, based on the fluorescence spectrum.

Description

A kind of phytoplankton composition quick determination method

Technical field

The present invention relates to floristics and/or quantity detection technique field, the phytoplankton composition quick determination method that uses in particularly a kind of marine eco-environment observation process.

Background technology

Phytoplankton in red tide identification determination techniques can roughly reduce two aspects: the determination techniques of 1. forming based on the phytoplankton of in-vivo analysis.The most classical the most frequently used phytoplankton classification and Detection method is to utilize optical microscope directly to observe form and the quantity of phytoplankton, and the method needs the professional to accomplish, and wastes time and energy, and can't satisfy the needs of on-the-spot fast monitored.Image recognition technique is on the microscope basis, to install CCD additional to videotape, and the phytoplankton picture-taking is got off, and discerns through processing and the image contrast in the image library, obtains the information of phytoplankton composition.Low cytometric analysis combines to have produced Flowcam technology with image technique, promptly let frustule be the individual layer state and in flow cell, flow, utilize the image of frustule in the CCD picked-up flow cell after, carry out image recognition, the commercialization of this technology.But the image technique recognition capability can receive the influence of sample solution turbidity, nor is the technology of a popularization and application.2. use chemical analysis technology, obtain the information that phytoplankton is formed.Research at present is based on the phytoplankton constitutes analysis method of phytoplankton pigment analysis comparatively widely.But the pigment analysis method needs high performance liquid chromatograph, and reagent consumption is big, and sample pre-treatments is loaded down with trivial details, has limited it in the business Application in Monitoring.The gene analysis technique of newly-developed is a characteristic DNA fragment of confirming the target phytoplankton through certain screening technique, through the compare of analysis of dna fragmentation, realizes the phytoplankton composition measuring to unknown sample.But this technology is to set up molecular probe to single kind algae, is applicable to the discriminance analysis of specified sea areas to limited kind of algae of initiation red tide, and the analysis of the phytoplankton composition that a large amount of algae kinds coexist before taking place for red tide is still had any problem.

The instrument that above-mentioned technology relied on is not the common instrument of existing marine monitoring base station, is difficult to realize fast, differentiate in real time the practical demand of the phytoplankton that causes red tide yet.Phytoplankton living body fluorescent technology then has These characteristics, and does not need pre-treatment to reach " green " characteristic that does not produce waste liquid before having sample determination.

Over past ten years, fluorescent technique is obtaining very big development aspect the composition monitoring of marine phytoplankton group.Kolbowski etc. (1995) have distinguished three main algae populations through initial fluorescence (F0); Beutler etc. (2002); Utilize phytoplankton live body chlorophyll fluorescence excitation spectrum, phytoplankton is divided into four big types of (green alga, blue-green algaes; Latent algae; Mixed algae (containing dinoflagellate and diatom)) identification is measured, and has set up phytoplankton composition measuring technology, Here it is commercial at present BBE algae analyser.But BBE algae analyser can't be with most important two types of algae diatoms in China offshore sea waters and dinoflagellate Classification and Identification, thereby is difficult to satisfy the needs of China's marine site phytoplankton monitoring.

Domestic the phytoplankton detection technique of fluorescence has also been carried out extensive studies; Zhao Dongzhi etc. are in 863 invention (modularization red tide satellite remote sensing monitoring technology; Under support 2001AA636020); Adopt field measurement and indoor cultivation dual mode to measure the red tide and the non-red tide algae curves of spectrum such as Nitzschia closterium minutissima, Dicrateria inornata, tower born of the same parents algae, flat algae and chlorella such as dinoflagellate, Heterosigma akashiwo, fork angle algae, ocean blue green alga. adopt the normalization fluorescence height method of measuring the chlorophyll fluorescence peak heights that the sun excites, set up the relation of different algal species normalization fluorescence height and chlorophyll concentration.Set up the relation of different algal species fluorescence peak position and chlorophyll-a concentration simultaneously.Gold sea otter etc. carries out the ratio spectral analysis and obtains containing chlorophyll b, the tangible distinguishing characteristics of chlorofucine marine alga, thereby has carried out significant trial for the marine alga vivo identification.In addition; They also carry out the derivative analysis of second of fluorescence excitation spectrum to single marine alga and mixing marine alga; Use the valley point and the peak point of second derivative fluorescence emission spectrum to be unique point; The normalization data of getting the relative intensity of the corresponding fluorescence emission spectrum of unique point is a characteristic parameter, mixes the linear multiple regression analysis of marine alga, realizes the fluorescence classification and determination of algae.

Above-mentioned these researchs and practical application have fully shown the prospect that the identification of red tide algae was measured when the living body fluorescent technology applied to phytoplankton population composition analysis and red tide takes place.But; Present research just be confined to certain type of phytoplankton with remarkable fluorescent characteristics mensuration or with phytoplankton rough be divided into several big type after discern monitoring, do not have a kind of phytoplankton fluorescence analysis of when red tide takes place, the red tide algae that causes red tide being differentiated yet.

(the main phytoplankton monoid in China offshore sea waters has diatom, dinoflagellate, green alga, xanthophyta, blue-green algae, chrysophyceae and latent algae to the phytoplankton structure characteristics, and group's composition always has comparative advantage with the algae of 1-2 class during according to China's coastal waters phytoplankton compositing characteristic and red tide generation; When red tide takes place; Mostly be the unipolar type red tide; The biphasic or bipolar type red tide is more rare, and with regard to China offshore sea waters, the algae kind that red tide takes place is limited and confirms relatively; Utilize fluorescent technique to differentiate that the phytoplankton that causes red tide is fully possible in the level that belongs to when therefore red tide takes place); Based on the phytoplankton fluorescence spectrum, to set up and a kind ofly measure phytoplankton on can level under " normally " (red tide does not take place) situation and form in class, the fluorescent identification method that the phytoplankton that red tide takes place was measured in identification on the level that belongs to when red tide took place is pressing for of China's immediate offshore area ecological monitoring.

Summary of the invention

Technical matters to be solved by this invention is; A kind of phytoplankton composition quick determination method is provided; Phytoplankton structure characteristics during according to China's coastal waters phytoplankton compositing characteristic and red tide generation; Based on the phytoplankton fluorescence spectrum, can when red tide does not take place, can on the level of class, measure phytoplankton and form, on the level that belongs to, differentiate when red tide takes place and measure the phytoplankton that causes red tide.The identification that present fluorescence analysis just is confined to certain type of phytoplankton with remarkable fluorescent characteristics measure or with phytoplankton rough be divided into several big type after discern mensuration; Particularly for China offshore sea waters very important diatom and dinoflagellate; The existing fluorescence analysis that comprises BBE all can't be with they classification and Detection, also do not have a kind of technology of can be when red tide takes place the algae kind that causes red tide being carried out scene, Real time identification monitoring.Phytoplankton structure characteristics during according to China's coastal waters phytoplankton compositing characteristic and red tide generation; Set up and a kind ofly measure phytoplankton on can level under " normally " (red tide does not take place) situation and form in class; On the level that belongs to, discerning the technology of measuring the phytoplankton that red tide takes place when red tide takes place is pressing for of China's immediate offshore area ecological monitoring; It also is most possible at present one of on-the-spot, quick, the real-time phytoplankton structure monitoring technology that realizes; The realization of this technology will be the research red tide life mechanism that disappears, and then the control red tide provides strong technical guarantee.

For solving the problems of the technologies described above, the invention provides a kind of phytoplankton composition quick determination method, may further comprise the steps:

(1) obtains phytoplankton fluorescence spectrum original spectrum storehouse;

(2) make up the higher-dimension orthogonal intersection space, phytoplankton fluorescence spectrum original spectrum storehouse is projected to this higher dimensional space, extract the phytoplankton fluorescence spectral characteristic, set up phytoplankton fluorescence spectral characteristic spectrum storehouse;

(3) phytoplankton fluorescent spectrum is carried out discriminance analysis, provide kind and/or the quantity of contained phytoplankton in institute's test sample article.

Said fluorescence spectrum can comprise: by the continuous three-dimensional fluorescence spectrum that fluorospectrophotometer obtains, and the discrete three-dimensional fluorescence spectrum that many fluorescence excitation spectrums appearance obtains.

The method of said structure higher-dimension orthogonal intersection space is preferably: make up the higher-dimension orthogonal intersection space through wavelet analysis technology.

Said wavelet analysis technology can comprise: many wavelet techniques and/or wavelet packet technology.

Said step (3) may further include:

(3) utilize multiple linear regression and non-negative least square to resolve and set up phytoplankton fluorescence discriminance analysis technology, phytoplankton fluorescent spectrum is carried out discriminance analysis, provide kind and/or the quantity of contained phytoplankton in institute's test sample article.

Said step (2) may further include:

(2) utilize the orthogonal wavelet function to make up the higher-dimension orthogonal intersection space; Phytoplankton fluorescence spectrum original spectrum storehouse is projected to this higher dimensional space; In each projector space, select stable and fluorescent characteristics section or characteristic segments combination high specificity; As phytoplankton fluorescent characteristics spectrum, through cluster analysis technique construction phytoplankton fluorescence spectral characteristic spectrum storehouse.

Said step (3) may further include:

(3) phytoplankton fluorescent spectrum is carried out discriminance analysis; Provide kind and/or the quantity of contained phytoplankton in institute's test sample article: when red tide does not take place, on the level of class, measure phytoplankton by multiple linear regression and form, on the level that belongs to, discern when red tide takes place and measure the phytoplankton that causes red tide.

Said by wavelet analysis technology structure higher-dimension orthogonal intersection space, can realize through the following steps:

Adopt Haar small echo, the little wave system of Daubechies or Meyer small echo, select DB7 and COIF2 as wavelet basis function;

The wavelet character spectrum is defined as the projection of original fluorescence spectrum at wavelet space, and the characteristic segments that is projected as the wavelet character spectrum on each wavelet space adopts the Bayes techniques of discriminant analysis, selects suitable feature section or its combination to compose as recognition feature as required.

Said non-negative least square method, can realize through following step:

The characteristic spectrum of setting up is:

f _k=(f _K1, f _K2..., f _Kn), wherein, f _kBe k class phytoplankton standard spectrum, f _KiI data points for k class phytoplankton characteristic spectrum;

Obtain the spectrum vector corresponding behind the sample determination with characteristic spectrum:

F=(F ₁, F ₂..., F _n), wherein, F _iI data points for sample;

Every kind of algae is to the contribution (a of fluorescence spectrum _k) obtain F=∑ a through taking off alignment property regression equation _kf _k, weight factor w _Ki:

$x^2(a_1,a_2,\·\·\·a_5)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\left(\frac{F_i-\underset{k=1}{\overset{5}{\mathrm{\Σ}}}{a}_k{f}_{\mathrm{ki}}}{\underset{k=1}{\overset{5}{\mathrm{\Σ}}}{a}_k{w}_{\mathrm{ki}}}\right)$

x ²Being total error, is a _kNonlinear function, use iterative program, at first give a of denominator _kInitialize (a _k=1), calculates a of molecule then _k, with new a _kValue is given denominator, calculates molecule a _k, so circulation is up to molecule a _kWith denominator a _kEquate; A when the iteration appearance of negative _kThe time, this a during next iteration _kNo longer participate in iteration as 0; w _KiBe through the standard deviation of each phytoplankton in red tide fluorescent characteristics spectrum in the data of different characteristic section _KiObtain: w _Ki=σ _Kt

In the said step (2); The step of setting up phytoplankton fluorescence spectral characteristic spectrum storehouse may further include: Rayleigh scattered band in the original three-dimensional spectroscopic data is removed through the Delaunay triangular interpolation method; Again the spectrum after handling is carried out wavelet decomposition successively; Select DB7 small echo second layer scale component and second and third layer Wavelet Component, the 3rd layer of scale component characteristic spectrum of COIF2 small echo and second and third layer Wavelet Component are as characteristic spectrum.

The beneficial technical effects of assay method of the present invention is: directly measure seawater sample, need not pre-treatment, and in several minutes, provide classification and the quantity result of main phytoplankton fast.With similar technology leading in the world---the fluorescence algae analyser of German bbe-moldaenke company is compared; The present invention can discern the swim alga (comprising diatom and dinoflagellate that CHINESE OFFSHORE is very important) of measuring seven classes, and recognition correct rate reaches more than 90%; And can measure the identification that single algae and two kinds of algae biased samples belong to level, recognition correct rate reaches more than 80%.

Description of drawings

Fig. 1 is the identifying synoptic diagram that the present invention implements the said phytoplankton fluorescence identification technique of grain;

Among Fig. 2, Fig. 2 a is that the present invention implements the said middle original fluorescence spectrum figure of Skeletonemacostatum of grain, and Fig. 2 b is the fluorescence spectrum after the present invention implements the said removal scattering of grain;

Fig. 3 is that the present invention implements the said 37 kinds of algae DB7 small echo second layer scale component characteristic spectrums of grain;

Fig. 4 is that the present invention implements said 37 kinds of second and third layer of the algae DB7 small echo Wavelet Component characteristic spectrums of grain;

Fig. 5 is that the present invention implements the said 37 kinds of algae COIF2 small echo second layer scale component characteristic spectrums of grain;

Fig. 6 is that the present invention implements said 37 kinds of second and third layer of the algae COIF2 small echo Wavelet Component characteristic spectrums of grain.

Embodiment

The invention discloses the fluorescence analysis method that a kind of phytoplankton in red tide group forms; Phytoplankton structure characteristics when this method takes place according to China's coastal waters phytoplankton compositing characteristic and red tide; Based on the phytoplankton fluorescence spectrum; That sets up a kind ofly measures phytoplankton on can the level in class under " normally " (red tide does not take place) situation and forms, when red tide takes place on the level that belongs to identification measure the fluorescent identification method of the phytoplankton of generation red tide.Its basic fundamental route is: gather common phytoplankton sociales of CHINESE OFFSHORE and red tide kind separation and Culture, obtain fluorescence spectrum, set up phytoplankton fluorescence spectrum original spectrum storehouse; Utilize wavelet analysis, bayes discriminant analysis methods then; Phytoplankton in red tide live body three-dimensional fluorescence spectrum is projected to constructed higher-dimension orthogonal intersection space; The characteristic segments that is projected as characteristic spectrum on each space; In each projector space, select stable and fluorescent characteristics section or characteristic segments combination high specificity; Form phytoplankton in red tide fluorescent characteristics spectrum,, compose the storehouse with this and be the basis through cluster analysis technique construction phytoplankton fluorescent characteristics spectrum storehouse; Can when red tide does not take place, on the level of class, measure phytoplankton by multiple linear regression foundation and form, on the level that belongs to, discern the fluorescent identification method of measuring the phytoplankton that causes red tide when red tide takes place.The phytoplankton fluorescent spectrum of this technology after to wavelet decomposition carries out discriminance analysis; Attempt all combinations of characteristic spectrum between different plant species in the characteristic spectrum storehouse; Select the minimum pairing regression coefficient of combination of residual error as analysis result, provide classification and the quantity of contained phytoplankton in institute's test sample article thus.

A kind of embodiment of the quick identification assay method that phytoplankton of the present invention is formed can may further comprise the steps:

(1) obtains phytoplankton fluorescence spectrum original spectrum storehouse;

(2) make up the higher-dimension orthogonal intersection space through wavelet analysis technology, the phytoplankton fluorescence spectrum is projected to this higher dimensional space, extract the phytoplankton fluorescence spectral characteristic, set up phytoplankton fluorescence spectral characteristic spectrum spectrum storehouse;

(3) utilize multiple linear regression to be aided with non-negative least square and resolve the phytoplankton fluorescence discriminance analysis technology of setting up;

(4) phytoplankton fluorescent spectrum is carried out discriminance analysis, provide kind and the quantity of contained phytoplankton in institute's test sample article.

This method is primarily aimed at fluorescence spectrum (by the continuous three-dimensional fluorescence spectrum of fluorospectrophotometer acquisition and the discrete three-dimensional fluorescence spectrum of other many fluorescence excitation spectrums appearance acquisitions), also comprises the application on other spectrum.

The present invention is that to one of contribution of prior art the method for described extraction phytoplankton fluorescence spectrum intrinsic characteristics is wavelet analysis technology (comprising that many wavelet techniques and wavelet packet are technological).

The present invention is to utilize the orthogonal wavelet function to make up higher dimensional space to two of the contribution of prior art; The phytoplankton fluorescence spectrum is projected to this higher dimensional space; In each projector space, select stable and fluorescent characteristics section or characteristic segments combination high specificity; As phytoplankton fluorescent characteristics spectrum, through cluster analysis technique construction phytoplankton fluorescent characteristics spectrum storehouse.

Assay method of the present invention is to be the basis composing the storehouse with this; Can when red tide does not take place, on the level of class, measure phytoplankton by multiple linear regression foundation and form, on the level that belongs to, discern the fluorescent identification method of measuring the phytoplankton that causes red tide when red tide takes place.

The present invention is phytoplankton structure characteristics when taking place according to China's coastal waters phytoplankton compositing characteristic and red tide; Based on phytoplankton living body fluorescent spectrum; Stoichiometries such as comprehensive utilization wavelet analysis, cluster analysis learn a skill; Make up phytoplankton fluorescent characteristics spectrum storehouse, set up corresponding phytoplankton fluorescent identification method.Its basic fundamental route: 1. collect CHINESE OFFSHORE phytoplankton sociales, red tide kind and separation and Culture; Use fluorospectrophotometer, under 400-600nm wavelength exciting light, measure the emission spectrum of its 600-800nm scope; After the density normalization, set up phytoplankton fluorescence spectrum original spectrum storehouse; 2. fully utilize Chemical Measurement technique construction high position data spaces such as wavelet analysis, cluster analysis then; Phytoplankton live body three-dimensional fluorescence spectrum is projected to constructed higher-dimension orthogonal intersection space; The characteristic segments that is projected as characteristic spectrum on each space; In each projector space, select stable and the fluorescent characteristics section of high specificity, extraction phytoplankton fluorescent characteristics spectrum is set up phytoplankton fluorescence spectral characteristic spectrum spectrum storehouse; 3. compose the storehouse with this and be the basis; Be based upon by multiple linear regression analysis technology and non-negative least square method and on the class level, measure phytoplankton when red tide does not take place and form; When taking place, red tide on the genus level, discerns the fluorescent identification method of the phytoplankton that causes red tide; Phytoplankton fluorescent spectrum is analyzed in practical application, confirmed whether red tide to take place through measuring chlorophyll content, thereby use corresponding fluorescent characteristics spectrum storehouse and discrimination method.

1. the separation of Marine Planktonic algae, purifying and cultivation

According to China offshore sea waters particularly red tide and the biological survey data in the East Sea, choose the modal 37 kinds of phytoplanktons can reflecting offshore sea waters phytoplankton ecological characteristic basically in offshore sea waters.To each algae of choosing; With reference to manual on harmful marine microalgae (2003; Published byUnited Nations Educational, Scientific and Culture Organization) standard method in, separation, purifying and training objective algae kind.

2. the foundation of phytoplankton fluorescence identification assay method

Phytoplankton structure characteristics during according to China's coastal waters phytoplankton compositing characteristic and red tide generation; Phytoplankton fluorescence identification technology realizes on two levels; Be on the level of class, the phytoplankton composition to be discerned mensuration under the normal ecologic regime in marine site; And when red tide takes place, on the level that belongs to, the phytoplankton that causes red tide is differentiated.In practical application, confirm whether red tide to take place through measuring chlorophyll content, thereby use corresponding fluorescent characteristics spectrum storehouse and discrimination method.

Technology such as present technique utilization wavelet analysis realize goal in research.

The foundation in A phytoplankton fluorescence standard spectrum storehouse

◆ wavelet analysis technology

Chemometrics methods such as this research application wavelet technique; Make up specific higher dimensional space; With the three-dimensional fluorescence spectrum projection in higher dimensional space; Through compare of analysis, the fluorescence component of selecting high specificity in each space and under the different ecological environmental baseline, having a good stability is formed phytoplankton in red tide fluorescent characteristics spectrum, on this basis; Utilize multiple linear regression and the foundation of non-negative least square technology when red tide does not take place, on the class level, to measure phytoplankton and form, on the genus level, discern the fluorescent identification method of the phytoplankton that causes red tide when red tide takes place.

The ultimate principle of wavelet decomposition is following:

Under the linear assumption, the fluorescence spectrum that (1) formula has provided multiple algae potpourri does

$P\left(f\right)=\underset{m=1}{\overset{M}{\mathrm{\Σ}}}C\left(m\right)H(m,f)---\left(1\right)$

Wherein P (f) is the fluorescence spectrum of multiple algae potpourri, H (m, f), (m=1,2 ... M, f=1,2 ... F) be the fluorescence spectrum of unit concentration algae, f is the measuring frequency point number, and m is algae kind numbering.

(1) window function W (f) is multiply by on the formula both sides simultaneously

$W\left(f\right)P\left(f\right)=\underset{m=1}{\overset{M}{\mathrm{\Σ}}}C\left(m\right)H(m,f)W\left(f\right)---\left(2\right)$

(2) formula is made multiscale analysis.Select suitable quadrature yardstick base

Its corresponding wavelet basis is ψ _{J, n}, establish Φ _jAnd ψ _jBe respectively j layer metric space and the wavelet space that constitutes by yardstick base and wavelet basis.Utilize the relation of metric space and wavelet space

①Φ _j⊥ψ _j (3)

② ${\mathrm{\Φ}}_{j+1}={\mathrm{\Φ}}_j\⊕{\mathrm{\Ψ}}_j---\left(4\right)$

(2) formula of expansion gets

Wherein, subscript j representes the number of plies of multiscale analysis,

$d_{k,n}=\underset{f=1}{\overset{F}{\mathrm{\Σ}}}W\left(f\right)P\left(f\right){{\mathrm{\ψ}}^{*}}_{k,n}\left(f\right),k=\mathrm{1,2},\·\·\·,j---\left(8\right)$

$e_{k,n}^m=\underset{f=1}{\overset{F}{\mathrm{\Σ}}}W\left(f\right)H(m,f){{\mathrm{\ψ}}^{*}}_{k,n}\left(f\right),k=\mathrm{1,2},\·\·\·,j,m=\mathrm{1,2},\·\·\·,M---\left(9\right)$

" * " representes conjugation, and subscript j and n represent that respectively yardstick stretches and the translation of scaling function along the f axle.

For the given yardstick contraction-expansion factor or the number of plies j of multiscale analysis; Yardstick base

and wavelet basis are quadratures, promptly satisfy

② $<{\mathrm{\&psi;}}_{j,n}\&CenterDot;{\mathrm{\&psi;}}_{j,n^{\&prime;}}>=\underset{f=1}{\overset{F}{\mathrm{\&Sigma;}}}{\mathrm{\&psi;}}_{j,n}\left(f\right){{\mathrm{\&psi;}}^{*}}_{j,n^{\&prime;}}\left(f\right)=\left\{\begin{array}{cc}0& n\&NotEqual;n^{\&prime;}\\ 1& n=n^{\&prime;}\end{array}\right.---\left(11\right)$

Can confirm by (3)-(12) formula

$a_{j,n}=\underset{m=1}{\overset{M}{\mathrm{\&Sigma;}}}C\left(m\right)b_{j,n}^m,n=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,F/2.---\left(13\right)$

$d_{k,n}=\underset{m=1}{\overset{M}{\mathrm{\&Sigma;}}}C\left(m\right)e_{k,n}^m,n=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,F/2,k=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,j---\left(14\right)$

A wherein _InBe scale component, d _KnIt is Wavelet Component.Because conversion is a quadrature, each measures between component uncorrelated, and we can choose a wantonly when resolving C (m) _JnOr d _KnIn some component resolve.Scale component (also being low frequency component) a _JnWhat reflect is the information (the low frequency component reflection is than the information of large scale) that measurement data P (f) goes up more measuring frequency point, Wavelet Component (also being high fdrequency component) d _KnWhat reflect is the information (the high fdrequency component reflection is than the information of small scale) that measurement data P (f) goes up less measuring frequency point.

According to the known concentration algae fluorescence spectrum H of unit (m, f), m=1,2 ... M, f=1,2 ... F, select suitable quadrature yardstick base that it is carried out multiscale analysis [(7) formula with (9) formula], obtain corresponding scale component and Wavelet Component.In scale component and Wavelet Component, choose the characteristic that some component composition and classifications are estimated.

Successful application wavelet analysis key depends on following parameters of choice: the 1. selection of wavelet basis function; 2. the selection of scale component and Wavelet Component.

1. the selection of wavelet basis function

For ensureing the orthogonality of decomposing, must adopt Orthogonal Wavelets, like Haar small echo, the little wave system of Daubechies, Meyer small echo etc.Appropriate Orthogonal Wavelets is decomposed can highlight the scale component of algae unit's concentration fluorescence spectrum and the maximum difference of Wavelet Component.The present invention selects DB7 and COIF2 wavelet basis function.

2. the selection of scale component and Wavelet Component

Owing to adopted the quadrature decomposition, made each scale component or Wavelet Component all satisfy linear relationship.This makes to have given only selects part scale component or Wavelet Component to come deconvolution to become possibility.Component or component combination that selection differences is bigger are come deconvolution, on the one hand can amplify difference, improve correct class probability, on the other hand can amount of compressed data, reduce calculated amount.

The i Feature Selection

The wavelet character spectrum is defined as the projection of original fluorescence spectrum at wavelet space, and the characteristic segments that is projected as the wavelet character spectrum on each wavelet space can select suitable feature section or its combination to compose as recognition feature in practical application as required.The present invention will adopt the Bayes techniques of discriminant analysis to select:

The Bayes techniques of discriminant analysis is a basic skills in the pattern-recognition; This method had both been considered the overall probability size that occurs of all kinds of references; Considered that again the loss that causes because of erroneous judgement is big or small; Be applicable at sample size insufficient greatly, carry out condition discrimination under the unfavorable situation of large sample statistical theory, be fit to this research use.The basic application principle of Bayes diagnostic method is following:

Suppose that na bar spectrum as training set is by m variable x ₁, x ₂..., x _mForm the a kind of phytoplankton A that all classifies from needs ₁, A ₂..., A _lFor other nx bar fluorescence Spectra, each bar spectrum is labeled as x=(x with vector ₁, x ₂..., x _m), need to judge from any phytoplankton.That is, regard x as in the m-dimensional space a point, the Bayes criterion is found out a space R exactly ^(m)Be divided into mutually disjoint 1 complete subspace R ₁, R ₂..., R _lA kind of division methods.In case after the subspace was delimited, fluorescence spectrum x just must drop on and only drop on a certain subspace R in this 1 complete subspace _gIn, then put x under g kind A _g, that is, determine the fluorescence spectrum which kind of phytoplankton x belongs to.In assorting process, utilize the calculating of " maximum likelihood analytical test " (likelihood ratio test) to obtain best classification results.This method can be through open-and-shut its misjudged probability and the ownership of seeing of mapping when providing a kind of differentiation accuracy of spectrum.

The present invention has selected DB7 small echo second layer scale component and second and third layer Wavelet Component as characteristic spectrum; Selected the 3rd layer of scale component characteristic spectrum of COIF2 small echo and second and third layer Wavelet Component as characteristic spectrum.

Utilize the screening of Bayes techniques of discriminant analysis and confirm phytoplankton fluorescent characteristics spectrum; Under study for action; We find; The phytoplankton fluorescent characteristics spectrum that the different orthogonal wavelet function is extracted different classes of phytoplankton is had different classification capacities, and these characteristic spectrums has very strong complementarity on recognition capability.Therefore, this research has been set up level Four spectrum storehouse according to recognition capability and complementarity.

Ii standard spectrum storehouse is set up

The phytoplankton fluorescence spectrum can receive the illumination of growing environment; Temperature and growth period etc. factor influence; Therefore the present invention will carry out the phytoplankton culture experiment under a plurality of conditions; Obtain the fluorescent characteristics spectrum under the different growth conditionss of reflection phytoplankton; But because the spectrum of the most of recognition features under the phytoplankton different condition of the same race is similar, and also be similar with most of recognition features spectrums of the different phytoplanktons of planting of class and type of belonging to together phytoplankton not of the same race, the similar features existence meeting of composing has a negative impact to identification in a large number; Therefore, need extract and can represent each class and each to belong to phytoplankton fluorescent characteristics and the least possible characteristic spectrum---phytoplankton fluorescence standard spectrum.The present invention adopts hierarchical clustering method to obtain class level of phytoplankton and the standard spectrum on the genus level.To carry out cluster analysis with class or all phytoplanktons of type of belonging to together fluorescent characteristics spectrum under different condition of culture, be that one type fluorescent characteristics spectrum is made even and all obtained the fluorescence standard spectrum to gathering.

Hierarchical clustering method is the most a kind of method of using in the cluster analysis; Its cluster principle is at first to regard the sample or the index of some as one type separately, and then per sample or the close and distant degree of index, two types that close and distant degree is the highest merge; Class after considering then to merge and the close and distant degree between other types; Merge again, repeat this process, can all samples be merged into one type at last.Cluster process can come figurative expression to come out with the clustering tree that is called pedigree chart.The present invention confirms the required number that gathers as required.

◆ many wavelet techniques and wavelet packet technology

Many small echos are meant many small echos that two or more functions generate as scaling function.Many small echos integrate the superperformance of symmetry, slickness, short supportive, orthogonality and high-order vanishing moment.The basic thought of many small echos is the multiresolution analysis spaces that generated by single scaling function in single small echo, expands to by a plurality of scaling functions to generate, and obtains bigger degree of freedom with this.Therefore the generation of multiresolution analysis and many small echos is closely-related, is one of important directions of the many small echos of structure.For single small echo, many small echos have stronger resolution characteristic, so aspect signal Processing, more have superiority.

In the multiresolution analysis of wavelet transformation; Signal has been carried out effectively successively decomposing; But every layer of decomposition all is that the low frequency signal that decompose on the upper strata is decomposed into low frequency and high frequency two parts again, HFS is not decomposed again, thereby the resolution of high band is relatively poor.Wavelet packet analysis is a kind of improvement to multiresolution analysis; It will do further to decompose less than the HFS of segmentation, signal decomposed in the full range band, and can carry out the selection of frequency band; Thereby be a kind of meticulousr signal analysis method, have using value widely.With regard to it extracts fluorescence spectral characteristic, can utilize the information of HFS more fully, thereby extract more meticulous information.

The utilization of many small echos and wavelet packet is similar to wavelet analysis technology, repeats no more at this.

B phytoplankton fluorescent identification method is set up

◆ non-negative least square method

On the basis that characteristic spectrum is established, set up corresponding identification assay method:

The characteristic spectrum of setting up does

f _k=(f _K1, f _K2..., f _Kn), wherein, f _kBe k class phytoplankton standard spectrum, f _KiI data points for k class phytoplankton characteristic spectrum.

Can obtain the spectrum vector corresponding behind the sample determination with characteristic spectrum:

F=(F ₁, F ₂..., F _n), wherein, F _iI data points for sample.

Every kind of algae is to the contribution (a of fluorescence spectrum _k) can obtain F=∑ a through taking off alignment property regression equation _kf _kYet,, in actual conditions, this additivity can receive the influence of phytoplankton species and SPECTRAL REGION and present inconsistency, when utilizing non-negative least square method to solve an equation, has introduced weight factor w for this reason _Ki:

$x^2(a_1,a_2,\&CenterDot;\&CenterDot;\&CenterDot;a_5)=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(\frac{F_i-\underset{k=1}{\overset{5}{\mathrm{\&Sigma;}}}{a}_k{f}_{\mathrm{ki}}}{\underset{k=1}{\overset{5}{\mathrm{\&Sigma;}}}{a}_k{w}_{\mathrm{ki}}}\right)$

x ²Being total error, is a _kNonlinear function.The non-linear min algorithm of this equation is ripe, promptly uses iterative program, at first gives a of denominator _kInitialize (a _k=1), calculates a of molecule then _k, with new a _kValue is given denominator, calculates molecule a _k, so circulation is up to molecule a _kWith denominator a _kEquate.When negative value appears in the result of calculation of the algae of certain class, (possibly be because this class concentration of algae is low, and receive noise effect), be i.e. a of iteration appearance of negative _kThe time, this a during next iteration _kNo longer participate in iteration as 0.w _KiBe through discreteness (standard deviation) σ of each phytoplankton in red tide fluorescent characteristics spectrum in the data of different characteristic section _KiObtain: w _Ki=σ _Ki

Further specify the present invention below in conjunction with accompanying drawing and through instantiation.

As shown in Figure 1; The specified operational procedure of this method is following: for a position sample; At first measure fluorescence spectrum, calculate chlorophyll content then, if chlorophyll content does not surpass the red tide threshold value; This fluorescent spectrum gets into class identification mensuration program, promptly uses class standard spectrum storehouse to carry out discriminance analysis; If chlorophyll content surpasses the red tide threshold value, this fluorescent spectrum at first gets into class identification mensuration program, obtains discerning the mensuration program in the genus level that gets into corresponding class behind the analysis result, promptly uses the genus standard spectrum storehouse of respective doors class to carry out discriminance analysis.

One, the phytoplankton fluorescence spectrum obtains

Phytoplankton sociales, the red tide kind phytoplankton (seeing table 1) in the multiple district of selected 37 kinds of East China Sea red tides.

37 kinds of phytoplanktons that table 1 experiment is selected for use

Phytoplankton is cultivated in the laboratory, and timing sampling carries out fluorescence spectral measuring, and every kind of algae measurement finishes, and carries out the spectroscopic assay of mixed algae again.Twice of the parallel sampling and measuring of each sample.Spectral range: excitation wavelength 400-600nm, emission wavelength 600-800nm, step-length 5nm excites slit 5nm, emission slit 5nm, sweep velocity 12000nm/min.

Two, the foundation in phytoplankton fluorescence spectrum spectrum storehouse

1. foundation---the wavelet analysis and the cluster analysis in extraction of phytoplankton fluorescence spectral characteristic and fluorescent characteristics spectrum storehouse

As shown in Figure 2, its horizontal ordinate is represented fluorescence exciting wavelength (EX), and ordinate is represented fluorescent emission wavelength (EM), and ordinate is represented fluorescence intensity.Original three-dimensional fluorescence spectrum is owing to contain scattering effect (shown in Fig. 2 a and Fig. 2 b) such as Rayleigh scattering; Covered the intrinsic characteristics of spectrum; Be not easy effectively to extract its spectral signature; Therefore at first Rayleigh scattered band in the original three-dimensional spectroscopic data etc. is removed through the Delaunay triangular interpolation method, again the spectrum after handling is carried out wavelet decomposition successively.Each Wavelet Component and scale component are as candidate feature spectrum.

Through bayes discriminatory analysis technology, the Classification and Identification ability of individual candidate feature spectrum is analyzed, selected DB7 small echo second layer scale component and second and third layer Wavelet Component as characteristic spectrum; Selected the 3rd layer of scale component characteristic spectrum of COIF2 small echo and second and third layer Wavelet Component as characteristic spectrum.

Shown in Fig. 3～6, each several part color showing value size among the figure, numerical value from small to large, color from dark blue to redness; Peak of curve representes that the fluorescence intensity here reaches high value among the figure, underestimate represent that then fluorescence intensity a little less than.

Test selected 37 kinds of phytoplanktons all from Chinese Marine University marine pollution chemistry in ecology laboratory.With the natural sea-water after filtering and nutrient solution according to the formulated nutrient culture media, and with the HCl solution adjusting pH of 1mol/L about 8.The thermograde of incubator is 20 ℃, and 25 ℃ of illumination gradients are 20000lux, 12000lux, 7000lux, 4000lux.The illumination exposure period is 12h: 12h.Two parts of every kind of parallel cultivations of algae.

The cultivation cycle of phytoplankton is 12d, whenever measures once at a distance from 48 hours with the Fluorolog3-11 fluorospectrophotometer that Jobin Yvon company produces.Sweep parameter is provided with: excitation wavelength 400～600nm, and emission wavelength 600～800nm, step-length is 5nm, slit width is 5nm.Sweep signal integral time is 0.05s.Twice of every duplicate samples replicate determination.Use is mixed by the chlorophyll concentration of the BBE fluorescence algae analyser measure portion sample of bbe moldaenke company production then in twos, makes the chlorophyll concentration of advantage algae account for 75%.Fig. 3 is 37 kinds of algae DB7 small echo second layer scale component characteristic spectrums.Fig. 4 is 37 kinds of second and third layer of algae DB7 small echo Wavelet Component characteristic spectrums.Fig. 5 is 37 kinds of algae COIF2 small echo second layer scale component characteristic spectrums.Fig. 6 is 37 kinds of second and third layer of algae COIF2 small echo Wavelet Component characteristic spectrums.

3. the foundation in phytoplankton standard fluorescence spectrum spectrum storehouse

Hierarchical clustering method is the most a kind of method of using in the cluster analysis; Its cluster principle is at first to regard the sample or the index of some as one type separately, and then per sample or the close and distant degree of index, two types that close and distant degree is the highest merge; Class after considering then to merge and the close and distant degree between other types; Merge again, repeat this process, can all samples be merged into one type at last.Cluster process can come figurative expression to come out with the clustering tree that is called pedigree chart.This patent is confirmed the required number that gathers as required.Obtain 189 DB7 scale component standard spectrums and 173 DB7 Wavelet Component standard spectrums altogether for 37 kinds of algaes, form 1 grade standard spectrum storehouse and the 2 grade standards spectrum storehouse of these algaes respectively.Obtain 179 COIF2 scale component standard spectrums and 191 COIF2 Wavelet Component standard spectrums, form 3 grade standards spectrum storehouse and the 4 standard spectrum storehouses of these algaes respectively.

The service routine in spectrum storehouse is: at first use 1 grade of spectrum storehouse to carry out discriminance analysis; If recognition result drops on the algae kind scope that can correctly discern in this spectrum storehouse; Then provide identification and measure the result,, then get into the spectrum storehouse that next stage can correctly be discerned this algae kind scope if recognition result drops on the algae kind scope that can not correctly discern in this spectrum storehouse; And the like, until providing the most believable result.

Three, phytoplankton fluorescence identification assay method

On the basis that characteristic spectrum is established, set up corresponding identification assay method:

The characteristic spectrum of setting up is:

f _k=(f _K1, f _K2..., f _Kn), wherein, f _kBe k class phytoplankton standard spectrum, f _KiI data points for k class phytoplankton characteristic spectrum.

Can obtain the spectrum vector corresponding behind the sample determination with characteristic spectrum:

F=(F ₁, F ₂..., F _n), wherein, F _iI data points for sample.

Every kind of algae is to the contribution (a of fluorescence spectrum _k) can obtain F=∑ a through taking off alignment property regression equation _kf _kYet,, in actual conditions, this additivity can receive the influence of phytoplankton species and SPECTRAL REGION and present inconsistency, when utilizing non-negative least square method to solve an equation, has introduced weight factor w for this reason _Ki:

$x^2(a_1,a_2,\&CenterDot;\&CenterDot;\&CenterDot;a_5)=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(\frac{F_i-\underset{k=1}{\overset{5}{\mathrm{\&Sigma;}}}{a}_k{f}_{\mathrm{ki}}}{\underset{k=1}{\overset{5}{\mathrm{\&Sigma;}}}{a}_k{w}_{\mathrm{ki}}}\right)$

x ²Being total error, is a _kNonlinear function.The non-linear min algorithm of this equation is ripe, promptly uses iterative program, at first gives a of denominator _kInitialize (a _k=1), calculates a of molecule then _k, with new a _kValue is given denominator, calculates molecule a _k, so circulation is up to molecule a _kWith denominator a _kEquate.When negative value appears in the result of calculation of the algae of certain class, (possibly be because this class concentration of algae is low, and receive noise effect), be i.e. a of iteration appearance of negative _kThe time, this a during next iteration _kNo longer participate in iteration as 0.w _KiBe through discreteness (standard deviation) σ of each phytoplankton in red tide fluorescent characteristics spectrum in the data of different characteristic section _KiObtain: w _Ki=σ _Ki

3. identification measurement result

The algae sample planted by more than 1500 lists that this technology is formed 37 kinds of phytoplanktons and the mixed algae sample can obtain 90% and 85% correct recognition rata respectively on the genus level, and the recognition correct rate on the class level remains on more than 95%; On genus and class level, can obtain 88.2% and 97.1% correct recognition rata to compound sample respectively.Can realize door, belong to the correct identification (obtaining 100% correct recognition rata) on the level the algae kinds such as Al, Pr, Pm, Ma, Gy, Gs, Sc, Km, Ps, Sk, Cl, Chaetoceros (Cu, De, Di), Tr, Ha, Cm, Is, Ks, PP, Ch, Pl, Ds, Pu, Mp, Cy, Rs and Ra as first sociales in the phytoplankton compound sample.

Claims (1)

1. phytoplankton composition quick determination method may further comprise the steps:

(1) obtains phytoplankton fluorescence spectrum original spectrum storehouse;

(2) make up the higher-dimension orthogonal intersection space, phytoplankton fluorescence spectrum original spectrum storehouse is projected to this higher-dimension orthogonal intersection space, extract the phytoplankton fluorescence spectral characteristic, set up phytoplankton fluorescence spectral characteristic spectrum storehouse;

(3) phytoplankton fluorescent spectrum is carried out discriminance analysis, provide kind and/or the quantity of contained phytoplankton in institute's test sample article;

Said fluorescence spectrum comprises: by the continuous three-dimensional fluorescence spectrum that fluorospectrophotometer obtains, and the discrete three-dimensional fluorescence spectrum that many fluorescence excitation spectrums appearance obtains;

The method of said structure higher-dimension orthogonal intersection space is: make up the higher-dimension orthogonal intersection space through wavelet analysis technology;

Said wavelet analysis technology comprises: many wavelet techniques and/or wavelet packet technology;

Said step (3) further comprises:

(3a) utilize multiple linear regression and non-negative least square method to resolve and set up phytoplankton fluorescence discriminance analysis technology, phytoplankton fluorescent spectrum is carried out discriminance analysis, provide kind and/or the quantity of contained phytoplankton in institute's test sample article;

Said step (2) further comprises:

(2a) utilize the orthogonal wavelet function to make up the higher-dimension orthogonal intersection space; Phytoplankton fluorescence spectrum original spectrum storehouse is projected to this higher-dimension orthogonal intersection space; In each projector space, select stable and fluorescent characteristics section or characteristic segments combination high specificity; As phytoplankton fluorescent characteristics spectrum, through cluster analysis technique construction phytoplankton fluorescence spectral characteristic spectrum storehouse;

Said step (3) further comprises:

(3b) phytoplankton fluorescent spectrum is carried out discriminance analysis; Provide kind and/or the quantity of contained phytoplankton in institute's test sample article: when red tide does not take place, on the level of class, measure phytoplankton by multiple linear regression and form, on the level that belongs to, discern when red tide takes place and measure the phytoplankton that causes red tide;

Said through wavelet analysis technology structure higher-dimension orthogonal intersection space, realize through the following step:

Adopt Haar small echo, the little wave system of Daubechies or Meyer small echo, select DB7 and COIF2 as wavelet basis function;

The wavelet character spectrum is defined as the projection of original fluorescence spectrum at wavelet space, and the characteristic segments that is projected as the wavelet character spectrum on each wavelet space adopts the Bayes techniques of discriminant analysis, selects suitable feature section or its combination to compose as recognition feature as required;

Said non-negative least square method, realize through following step:

The characteristic spectrum of setting up is:

f _k=(f _K1, f _K2..., f _Kn), wherein, f _kBe k class phytoplankton standard spectrum, f _KiI data points for k class phytoplankton characteristic spectrum;

Obtain the spectrum vector corresponding behind the sample determination with characteristic spectrum:

F=(F ₁, F ₂..., F _n), wherein, F _iI data points for sample;

Every kind of algae obtains F=∑ a to the contribution ak of fluorescence spectrum through taking off alignment property regression equation _kf _k, weight factor w _Ki:

$x^2(a_1,a_2,...a_5)=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(\frac{F_i-\underset{k=1}{\overset{5}{\mathrm{\&Sigma;}}}{a}_k{f}_{\mathrm{ki}}}{\underset{k=1}{\overset{5}{\mathrm{\&Sigma;}}}{a}_k{w}_{\mathrm{ki}}}\right)$

x ²Being total error, is a _kNonlinear function, use iterative program, at first give a of denominator _kInitialize a _k=1, calculate a of molecule then _k, with new a _kValue is given denominator, calculates molecule a _k, so circulation is up to molecule a _kWith denominator a _kEquate; A when the iteration appearance of negative _kThe time, this a during next iteration _kNo longer participate in iteration as 0; w _KiBe through the standard deviation of each phytoplankton in red tide fluorescent characteristics spectrum in the data of different characteristic section _KiObtain: w _Ki=σ _Ki

In the said step (2); The step of setting up phytoplankton fluorescence spectral characteristic spectrum storehouse further comprises: Rayleigh scattered band in the original three-dimensional spectroscopic data is removed through the Delaunay triangular interpolation method; Again the spectrum after handling is carried out wavelet decomposition successively; Select DB7 small echo second layer scale component and second and third layer Wavelet Component, the 3rd layer of scale component characteristic spectrum of COIF2 small echo and second and third layer Wavelet Component are as characteristic spectrum.

Images