Fluorescence spectrum analysis method for reducing algae misrecognition rate based on spectrum similarity

Abstract

The invention discloses a fluorescence spectrum analysis method for reducing an algae misrecognition rate based on spectrum similarity, belongs to the technical field of marine water ecological environment monitoring, and can effectively reduce the algae spectrum misrecognition rate. Compared with a standard method, the classification measurement result of 95 pure algae samples by the algorithm provided by the invention has four difference value points outside a 95% consistency limit, and the proportion is less than 5%. Compared with the national standard method, the classified measurement results of 49 algae samples in the mixed sample have three difference points outside the 95% consistency limit, and the proportion is also less than 5%. Considering the randomness of the measurement error, the result shows that the measurement result of the algorithm and the measurement result of the national standard method have good consistency.

Description

Technical Field

[0001] The present invention belongs to the technical field of marine water ecological environment monitoring, and in particular relates to a fluorescence spectrum analysis method for reducing the misidentification rate of algae based on spectral similarity. Background Art

[0002] Algae is a general term for a class of microscopic algae that live in the water in a planktonic manner. Algae are the material and energy basis of aquatic ecosystems. The composition, community structure and abundance of algae directly affect the water quality, energy flow, material flow and biological resource changes of the ecosystem.

[0003] The in vivo fluorescence method has the advantages of fast measurement speed, high sensitivity, and no need for sample pretreatment. It is currently the most promising method for rapid on-site measurement of phytoplankton concentration in large areas of water. The basic idea of ​​the in vivo fluorescence method is to invert the chlorophyll a concentration of algae through the fluorescence intensity generated by stimulated radiation of chlorophyll a molecules in living algal cells. The photosynthetic pigment composition of different types of living algal cells has interspecies differences, and therefore has its own characteristic fingerprint fluorescence spectrum. When there are multiple components in the sample, within the range of Lambert-Beer's law, the algae fluorescence spectrum satisfies linear additivity, so the quantitative problem of algae concentration can be transformed into a linear regression problem of the matrix. This is the principle basis for the rapid classification and quantitative measurement of phytoplankton through in vivo fluorescence spectroscopy.

[0004] Considering the practical significance and accuracy requirements of the analysis results, the spectrum is often analyzed based on the idea of ​​non-negative least squares multivariate linear regression analysis. However, there are many interferences in the background of natural water bodies, and the fluorescence of algae is somewhat unstable. The construction method of the standard spectral library of phytoplankton based on interpolation can significantly solve the following problems: 1. The problem of species misidentification caused by the deformation of the three-dimensional fluorescence spectrum of the same phytoplankton under different habitat conditions; 2. The problem of quantitative error of pigment concentration caused by differences in fluorescence efficiency. The method of establishing a standard spectral library by interpolation significantly improves the inversion accuracy of the measurement results of multivariate regression analysis, but it also brings new problems: the use of the interpolation standard spectral library has led to a surge in the combination of possible solutions, resulting in the solution with the smallest error when the least squares idea is used to analyze the aliased spectra of multi-component phytoplankton. It is not necessarily the optimal solution. This may cause low-concentration misdetection of non-target algae samples, which will have an adverse effect on the accurate determination of chlorophyll a concentration of phytoplankton. Summary of the invention

[0005] In view of the problem of spectral misidentification raised in the background technology, the present invention proposes a phytoplankton aliasing fluorescence spectrum feedback analysis algorithm based on spectral similarity judgment, which can effectively reduce the misidentification rate of algae spectra.

[0006] The technical solution of the present invention is as follows:

[0007] A fluorescence spectrum analysis method for reducing the misidentification rate of algae based on spectral similarity comprises the following steps:

[0008] Step 1: The unknown aliased three-dimensional fluorescence spectrum to be analyzed According to the idea of ​​non-negative least squares, global optimization is performed to obtain the initial solution combination. ( ), ( ) and residual .

[0009] Step 2: Initial unknown spectrum The three-dimensional fluorescence spectrum of the net algae after analysis After dimensionality reduction, the algae fluorescence spectra after dimensionality reduction are recorded as and The dimensionality reduction method is as follows: the excitation fluorescence spectrum of the three-dimensional fluorescence spectrum is expanded, and the first order is connected in the order of increasing emission wavelength to form a two-dimensional excitation fluorescence spectrum sequence. Among them: .

[0010] Step 3: and Input the first judgment layer, calculate its correlation coefficient value, and determine whether it meets the loop termination condition When the termination condition of the first judgment layer is not met, discard this group of solutions; and return to step 1 and re-analyze until the termination condition of the first judgment layer is met. Among them:

[0011] ,

[0012] ,

[0013] ,

[0014] ,

[0015] express and Correlation coefficient; express and Covariance; and Respectively and The variance of and After dimensionality reduction and The spectrum The number of spectral points is n and m represents the number of excitation wavelengths and emission wavelengths in the three-dimensional fluorescence spectrum, respectively.

[0016] Step 4: When the termination condition of the first judgment layer is met, enter the second judgment layer. Use the spectral dimension reduction method described in step 2 to calculate the residuals Standard spectrum of phytoplankton ( ) are processed and recorded as and ( ).

[0017] Step 5: Calculate separately and The correlation coefficient value of is used to determine whether it meets the termination condition of the second judgment layer. , when the termination condition of the second judgment layer is not met, discard this group of solutions and return to step 1 to re-analyze until the termination condition of the second judgment layer is met. Among them:

[0018] ,

[0019] ,

[0020] ,

[0021] ,

[0022] express and The correlation coefficient of express and Covariance; and Respectively and The variance of and express and Middle Spectral point value.

[0023] Step 6: Output the minimum non-negative solution that satisfies all judgment layers, denoted as ( ) and residual .

[0024] In the above technical solution, in step 1, the unknown aliased three-dimensional fluorescence spectrum to be analyzed Obtained through the following methods:

[0025] (1) Establishment of a standardized spectral library of different phyla of phytoplankton

[0026] Step 1: Obtain the fluorescence spectra of living phytoplankton of different classes, with an excitation wavelength range of 370 nm to 650 nm and an emission wavelength range of 600 nm to 720 nm. Measure the standard chlorophyll a concentration of the living phytoplankton;

[0027] Step 2: Dividing the living phytoplankton fluorescence spectrum obtained in step 1 by the standard chlorophyll a concentration of the living phytoplankton to obtain the standard spectra of the phytoplankton of different classes;

[0028] Step 3: Calculate the mean spectrum of the standard spectrum of the phytoplankton class according to the standard spectrum in step 2 and the corresponding relative standard deviation S;

[0029] Step 4: Calculation ± 2 S, to obtain the upper and lower limits of the interpolation range of the standard spectral library of phytoplankton of the said phylum;

[0030] Step 5: Perform uniform interpolation within the interpolation range of the phytoplankton standard spectrum library described in step 4, and the number of interpolations is 20;

[0031] Step 6: According to steps 1 to 5, construct a three-dimensional fluorescence standard spectral library of different types of phytoplankton;

[0032] (2) Obtain a three-dimensional fluorescence spectrum of the unknown phytoplankton sample in vivo, with the spectral range being the same as step 1 in 1.

[0033] (3) The spectrum is analyzed based on the idea of ​​non-negative least squares multivariate linear regression analysis.

[0034] In the above technical solution, the analytical method is as follows: for a certain type of algae, its three-dimensional fluorescence spectrum measurement value is a fluorescence intensity data matrix, theoretically:

[0035] ,

[0036] Among them, M i is the in vivo three-dimensional fluorescence measurement spectrum of a certain algae sample, c i represents the concentration of algae samples of this phylum, f i Represents the standardized fluorescence spectrum of algae samples of this phylum.

[0037] When there are multiple components in the sample, within the range of Lambert-Beer's law, the algae fluorescence spectrum satisfies the linear additivity, that is:

[0038] ,

[0039] ,

[0040] Where: M represents the aliased living three-dimensional fluorescence measurement spectrum of algae to be analyzed; f ij is the jth standard spectrum in the standard spectrum library of phytoplankton of the i-th phylum; c i is the chlorophyll a concentration of the i-th phylum, χ is the error; λ mn is the fluorescence intensity of the i-th algae at the n-th excitation wavelength and the m-th emission wavelength.

[0041] ,

[0042] In the above technical solution, there are many interferences in the natural water background, and the fluorescence of algae is unstable. The existence of background interference and the use of interpolation standard spectral library make the possible solution combinations increase sharply, resulting in the solution with the smallest error not necessarily being the optimal solution when the least squares idea is used to analyze the aliased spectra of multi-component phytoplankton. This will cause a certain probability of misidentification of species and have an adverse effect on the accurate determination of chlorophyll a concentration of phytoplankton.

[0043] Beneficial effects:

[0044] The fluorescence spectrum analysis method proposed in the present invention for reducing the misidentification rate of algae based on spectral similarity can effectively reduce the misidentification rate of algae spectra. Specifically, it has been confirmed through experiments that the classification measurement results of 95 pure algae samples by the algorithm of the present invention have 4 difference points outside the 95% consistency limit compared with the standard method, accounting for less than 5%. The classification measurement results of 49 algae samples in the mixed samples have 3 difference points outside the 95% consistency limit compared with the national standard method, also accounting for less than 5%. Taking into account the randomness of the measurement error, this result shows that there is a good consistency between the measurement results of the algorithm of the present invention and the measurement results of the national standard method. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a logical diagram of the feedback analysis algorithm for aliased fluorescence spectra of phytoplankton based on spectral similarity judgment.

[0046] Figure 2 This is a consistency analysis between the classification measurement results of pure algae samples by the algorithm described in the present invention and the standard method.

[0047] Figure 3 This is a consistency analysis between the classification measurement results of mixed algae samples by the algorithm described in the present invention and the standard method. DETAILED DESCRIPTION

[0048] The specific implementation mode of the present invention is described in detail below. This embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation mode and a specific operation process are given, but the protection scope of the present invention is not limited to the following embodiments.

[0049] Example 1 Three-dimensional fluorescence spectra of living phytoplankton under different habitat conditions

[0050] Common phytoplankton species in my country's coastal and freshwater waters were selected, including Cyanobacteria, Chlorophyta, Bacillariophyta, Dinophyta, Xanthophyta and Cryptophyta, see Table 1. They were cultured according to GB / T21805-2008 "Chemicals - Algal Growth Inhibition Test".

[0051] Table 1 Example of selected algae species

[0052]

[0053] Take an appropriate amount of algae sample mother solution at different growth stages, dilute it according to a certain volume ratio, and measure the three-dimensional fluorescence spectrum of the sample in vivo with a fluorescence spectrophotometer. The excitation wavelength range is 370 nm to 650 nm, and the emission wavelength range is 600 nm to 720 nm. At the same time, measure the standard concentration of chlorophyll a in the algae sample.

[0054] Example 2 Obtaining interpolation standard spectral libraries of different phytoplankton

[0055] The algae samples with known chlorophyll a standard concentration obtained in Example 1 were divided into an experimental set and a test set at a ratio of 7:3. According to the first part of the algorithm content described above, the three-dimensional fluorescence spectra of the phytoplankton living body in the experimental set were divided by the respective chlorophyll a standard concentrations, thereby obtaining a series of three-dimensional fluorescence standard spectra of the phytoplankton living body.

[0056] Calculate the mean fluorescence spectrum of the phytoplankton according to the standard spectrum of different categories and the corresponding relative standard deviation S; calculate ± 2 S, obtain the upper and lower limits of the interpolation range of the standard spectrum library of phytoplankton; perform uniform interpolation within the interpolation range of the standard spectrum library of phytoplankton described in step 4, the number of interpolations is 20, so as to establish the standard spectrum library of phytoplankton.

[0057] The above method obtained the standardized spectral library of phytoplankton of Cyanobacteria, Chlorophyta, Bacillariophyta, Dinophyta, Xanthophyta and Cryptophyta.

[0058] Example 3 Analysis effect evaluation parameters

[0059] The analytical method of the present invention is used to analyze the living fluorescence spectrum of the phytoplankton in the test set obtained in Example 1, and the analytical results are compared with the chlorophyll a standard concentration of the sample and the analytical results of the commonly used non-discriminatory least squares multivariate linear regression algorithm (hereinafter referred to as the original analytical algorithm). The test set samples include 96 pure algae samples (17 Cyanobacteria, 17 Chlorophyta, 21 Bacillariophyta, 17 Dinophyta, 14 Cryptophyta and 10 Xanthophyta), and 24 mixed samples.

[0060] The false recognition rate , false detection rate and relative measurement error Indicates the parsing accuracy of the algorithm.

[0061] False recognition rate It indicates the ratio of misidentified samples to the total number of samples when identifying a series of phytoplankton samples. Misidentification refers to the incorrect interpretation of the dominant population of the sample (e.g., Cyanobacteria are identified as non-Cyanobacteria or mixed algae samples with non-Cyanobacteria as the dominant population, or mixed algae samples with Cyanobacteria as the dominant population are identified as mixed algae samples with non-Cyanobacteria as the dominant population).

[0062] False positive rate It indicates the ratio of the chlorophyll a concentration of unrelated algae detected in the non-misidentified sample to the total chlorophyll a concentration detected in the sample when analyzing a specific algae sample.

[0063] Relative measurement error It indicates the ratio between the detected chlorophyll a concentration of a specific algae sample and the standard value of the chlorophyll a concentration of the sample when analyzing the specific algae sample.

[0064] ,

[0065] ,

[0066] ,

[0067] in: Represents the total number of samples in a series of phytoplankton of a particular phylum; It indicates the number of samples misidentified in the series of samples of phytoplankton of this specific phylum; It indicates the standard concentration of chlorophyll a of a certain phylum of phytoplankton sample; It represents the analyzed chlorophyll a concentration of a specific phylum of phytoplankton sample; Indicates the chlorophyll a concentration of other algae that were misidentified when analyzing a specific algae sample.

[0068] Example 4 Analysis Effect Evaluation

[0069] The analysis results are shown in Table 2-9 below. To simplify the description, A, B, H, K, M and X in the table represent cyanobacteria, green algae, diatoms, dinophyta, cryptophyta and yellow algae respectively; blanks represent "concentration is 0" or "not detected". In order to accurately test the classification and measurement ability of AFA under different algae biomass concentration levels, the algae sample concentrations were divided into low concentration groups (0~100 μg·L -1 ), medium concentration group (100-200 μg·L -1 ) and high concentration group (>200 μg·L -1 ).

[0070] Table 2 Comparison of measurement results of pure algae samples of Cyanobacteria

[0071]

[0072] Table 3 Comparison of measurement results of pure algae samples of Chlorophyta

[0073]

[0074] Table 4 Comparison of measurement results of pure algae samples of Diatomaceous phylum

[0075]

[0076] Table 5 Comparison of measurement results of pure algae samples of Dinophyta

[0077]

[0078] Table 6 Comparison of measurement results of pure algae samples of Cryptophyta

[0079]

[0080] Table 7 Comparison of measurement results of pure algae samples of Xanthophyta

[0081]

[0082] Table 8 Standard concentration of mixed algae samples

[0083]

[0084] Table 9 Comparison of measurement results of mixed algae samples

[0085]

[0086] Tables 2 to 7 show that when analyzing the fluorescence spectra of 16 pure phytoplankton living algae in the test set, the original analysis method and the spectrum analysis algorithm described in the present invention have a recognition accuracy of 100% for all pure samples. The original analysis method analyzed the chlorophyll a concentration of other algae in 6 algae samples with lower concentrations, while the algorithm described in the present invention only misidentified the lower concentration of Cryptophyta in 3 samples of the medium and high concentration groups, with a misidentification rate of 100%. The false positive rate of the three samples that were misidentified was reduced by 18.8%. They are 4.7%, 2.8%, and 1.1% respectively, with an average of 2.9%. It ranges from 0.2% to 20.6%, with an average of 8.1%.

[0087] When measuring the 17 pure samples of Chlorophyta in the test set, the spectral analysis algorithm of the present invention and the original analysis method have a recognition accuracy of 100% for all pure samples. The original analysis method detected one low-concentration algae from other phyla for each of 5 samples, and two low-concentration algae from other phyla for 2 samples, with a false positive rate of 100%. The variation range is 0.5%~5.4%, and the relative error of the measurement results is in the range of 2.2%~40.8%, with an average of 21.7%. In comparison, the algorithm of the present invention misdetected low concentrations of other algae in 6 samples of the low concentration group, with a false detection rate of 0.1%~1.6% respectively, with an average of 0.2%. It ranges from 0.9% to 37.9%, with an average of 15.3%.

[0088] When measuring 21 pure samples of Bacillariophyta, the original analysis algorithm had one misidentification, and the misidentification rate was The recognition accuracy of the algorithm described in the present invention is 100%. In addition, the original analysis method detected low concentrations of other algae in 10 samples among the correctly identified samples, with a false positive rate of 1. The variation range is 0.6%~37.1%, with an average of 12.2%; the relative error of measurement In the range of 0.4% to 100%, the average is 15.3%. In comparison, the algorithm of the present invention only misdetected low concentrations in 2 samples of the low concentration group, and the false detection rate was They are 0.6% and 13.7% respectively. It ranges from 0.2% to 30.4%, with an average of 14.1%.

[0089] When measuring 17 pure samples of Dinophyta, the original analysis algorithm had 2 misidentifications, with a misidentification rate of The accuracy of the algorithm described in the present invention is 100%. In addition, the false positive rate of the original analysis method for correctly identified samples is The variation range is 0.1%~24.8%, with an average of 10.6%; the relative error of measurement In the range of 0.2% to 100%, the average is 22.9%. In comparison, the algorithm of the present invention misdetected low-concentration algae of other phyla in the high, medium and low concentration groups. The misdetection ratio was high in the low concentration group and low in the medium and high concentration groups. The relative error is between 0.1% and 15.9%, with an average of 4.8%. It ranges from 0% to 19.1%, with an average of 8.1%.

[0090] When measuring 14 pure samples of Cryptophyta, the spectral analysis algorithm of the present invention and the original analysis method had a 100% recognition accuracy for all pure samples. The original analysis method detected low concentrations of other algae in 7 of the samples, but the spectral analysis algorithm of the present invention did not make any false detections for all samples in the high, medium and low concentration groups.

[0091] When measuring 10 pure samples of Xanthophyta, the spectral analysis algorithm of the present invention and the original analysis method have a 100% recognition accuracy for all pure samples. The original analysis method detected low concentrations of other algae in 5 of the correctly identified samples, with a false positive rate of The variation range is 1.4%~62.4%, with an average of 14.6%; the relative error of measurement The range was 4.3% to 66.5%, with an average of 23.6%. In comparison, the algorithm of the present invention only misdetected low concentrations in 3 samples, with a false positive rate of The variation range is 0.8%~4.2%, with an average of 2.3%. It ranges from 2.8% to 30.2%, with an average of 13.2%.

[0092] Tables 8 and 9 show that when measuring the 24 mixed algae samples in the test set, the classification and recognition accuracy of the spectral analysis algorithm of the present invention for the mixed samples is 100%. Generally, the measurement error of the algae samples in the medium concentration group is relatively small, while the error is relatively large when measuring the algae samples in the high and low concentration groups, with an average of about 6.7% of measurement misidentification.

[0093] The Bland-Altman consistency test was used to analyze the consistency of the classification measurement results of pure and mixed algae samples. The results are as follows Figure 2~3 As shown in the figure, BFM Indicates the measurement result of the chlorophyll a concentration of the sample by the algorithm of the present invention; STD It represents the chlorophyll a concentration measured by the national standard method. The deviation mean d represents the average value of the difference between the measured values ​​of the sample by the two methods; s represents the standard deviation of the difference between the measured values ​​of the sample by the two methods.

[0094] The above description is only a specific implementation of the present application, so that those skilled in the art can understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest range consistent with the principles and novel features applied for herein.

Claims

1. A fluorescence spectrum analysis method for reducing the misidentification rate of algae based on spectral similarity, characterized in that: The steps include: Step 1: The unknown aliased three-dimensional fluorescence spectrum to be analyzed According to the idea of ​​non-negative least squares, global optimization is performed to obtain the initial solution combination. ( ), ( ) and residual ; Step 2: The three-dimensional fluorescence spectrum of the net algae after analysis After dimensionality reduction, the algae fluorescence spectra after dimensionality reduction are recorded as and ; Step 3: and Input the first judgment layer, calculate its correlation coefficient value, and determine whether it meets the loop termination condition ; When the termination condition of the first judgment layer is not met, this group of solutions is discarded; And return to step 1 to re-analyze until the termination condition of the first judgment layer is met, where: ， ， ， ， express and Correlation coefficient; express and Covariance; and Respectively and The variance of and After dimensionality reduction and The spectrum The number of spectral points; n and m represent the number of excitation wavelengths and emission wavelengths in the three-dimensional fluorescence spectrum, respectively; Step 4: When the termination condition of the first judgment layer is met, enter the second judgment layer and use the spectral dimension reduction method described in step 2 to calculate the residuals. Standard spectrum of phytoplankton ( ) are processed and recorded as and ( ); Step 5: Calculate separately and The correlation coefficient value of is used to determine whether it meets the termination condition of the second judgment layer. , when the termination condition of the second judgment layer is not met, discard this group of solutions and return to step 1 for re-analysis until the termination condition of the second judgment layer is met, where: ， ， ， ， express and The correlation coefficient of express and Covariance; and Respectively and The variance of and express and Middle Spectral point values; Step 6: Output the minimum non-negative solution that satisfies all judgment layers, denoted as ( ) and residual .

2. The fluorescence spectrum analysis method for reducing the misidentification rate of algae based on spectral similarity according to claim 1 is characterized in that: In step 1, the unknown aliased three-dimensional fluorescence spectrum to be analyzed Obtained through the following methods: (1) Establishment of a standardized spectral library of different types of phytoplankton Step 1: obtaining the fluorescence spectra of living phytoplankton of different classes, with an excitation wavelength range of 370 nm to 650 nm and an emission wavelength range of 600 nm to 720 nm, and measuring the standard chlorophyll a concentration of the living phytoplankton; Step 2: Dividing the living phytoplankton fluorescence spectrum obtained in step 1 by the standard chlorophyll a concentration of the living phytoplankton to obtain the standard spectra of the phytoplankton of different classes; Step 3: According to the standard spectrum in step 2, calculate the mean spectrum N and the corresponding relative standard deviation S of the standard spectrum of the phytoplankton class; Step 4: Calculate N ± 2 S, to obtain the upper and lower limits of the interpolation range of the standard spectral library of phytoplankton of the said phylum; Step 5: Perform uniform interpolation within the interpolation range of the phytoplankton standard spectrum library described in step 4, and the number of interpolations is 20; Step 6: According to steps 1 to 5, construct a three-dimensional fluorescence standard spectral library of different types of phytoplankton; (2) Obtaining a three-dimensional fluorescence spectrum of the unknown phytoplankton sample in vivo, the spectral range is the same as step 1 in 1; (3) The spectrum is analyzed based on the idea of ​​non-negative least squares multivariate linear regression analysis. The analysis method is as follows: for a certain type of algae, its three-dimensional fluorescence spectrum measurement value is a fluorescence intensity data matrix, theoretically: ， Among them, M i is the in vivo three-dimensional fluorescence measurement spectrum of a certain algae sample, c i represents the chlorophyll a concentration of the algae sample of this phylum, f i Indicates the standardized fluorescence spectrum of algae samples of this phylum; When there are multiple components in the sample, within the range of Lambert-Beer's law, the algae fluorescence spectrum satisfies the linear additivity, that is: ， ， ， Where: M represents the aliased living three-dimensional fluorescence measurement spectrum of algae to be analyzed; f ij is the jth standard spectrum in the standard spectrum library of phytoplankton of the i-th phylum; c i is the chlorophyll a concentration of the i-th phylum, χ is the error; λ mn is the fluorescence intensity of the i-th algae at the n-th excitation wavelength and the m-th emission wavelength.

3. The fluorescence spectrum analysis method for reducing the misidentification rate of algae based on spectral similarity according to claim 1, characterized in that: In step 2, the dimension reduction method is: the excitation fluorescence spectrum of the three-dimensional fluorescence spectrum is expanded, and the first order is connected in the order of increasing emission wavelength to form a two-dimensional excitation fluorescence spectrum sequence.

Images