A method for real-time in-situ analysis of biochemical components of microalgae by combining Raman spectroscopy detection and neural network model
By combining Raman spectral detection and neural network model, a one-dimensional convolutional neural network model is constructed, which solves the real-time and in-situ problems of microalgae biochemical component detection, and realizes efficient and accurate analysis of microalgae biochemical component.
Patent Information
- Application Number
- CN202411154482.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-22
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-08-22
AI Technical Summary
The existing technology cannot detect microalgae biochemical components in real time and in situ. The traditional methods take a long time, are not in situ and cumbersome, and cannot meet the real-time, accurate and rapid detection needs of important microalgae organic matter in the production process.
Combining Raman spectral detection and neural network model, by obtaining Raman spectral data and target variable standard values of microalgae samples, pre-processing and constructing a quantitative component detection model, using one-dimensional convolutional neural network for training and verification, to achieve the Real-time prediction of dry weight and fucoxanthin yield of microalgae cells.
It realizes efficient, real-time and in-situ detection of microalgae biochemical components, which can accurately predict key production parameters of microalgae cell populations, and improves the accuracy of detection and multi-parameter prediction capabilities.
Smart Images

Figure CN118688179B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of biochemical detection technology, and in particular to a method for real-time in-situ analysis of biochemical components of microalgae by combining Raman spectroscopy detection and a neural network model. Background Art
[0002] Microalgae are a type of eukaryotic or prokaryotic microorganisms that can perform photosynthesis. They have received widespread attention as a green and sustainable biological resource. In order to improve the utilization efficiency and yield of microalgae, it is particularly important to quantitatively analyze their cell active substances efficiently and in situ. However, the tiny size of microalgae has brought great difficulties to the detection of its physiological indicators. Traditional methods for detecting biochemical components of microalgae are mostly based on empirical evaluation or chemical analysis, which have the problems of being time-consuming, non-in situ, and cumbersome processes. They cannot meet the needs of real-time, accurate, and rapid detection of important organic matter in microalgae during the production process.
[0003] In order to improve the detection efficiency, a method for detecting microalgae biochemical components based on Raman spectroscopy was proposed. Raman spectroscopy is a spectral analysis method based on molecular vibration information, which can greatly improve the quantitative detection efficiency of target compounds. However, due to the interference of nonlinear factors such as strong chlorophyll autofluorescence and overlapping Raman spectral peaks in microalgae cells, the application of this technology in the field of quantification of intracellular compounds in microalgae is currently limited to the single-cell level, and needs to be combined with microfluidics, optical tweezers and microscopes, which is not conducive to the efficient, real-time and in situ detection of microalgae biochemical components. Summary of the invention
[0004] Based on the above application requirements and technical background, in order to solve the technical problem that the existing technology cannot detect the biochemical components of microalgae in real time and in situ, this application adopts the following technical solution:
[0005] In a first aspect, the present application proposes a method for real-time in-situ analysis of biochemical components of microalgae by combining Raman spectroscopy detection and a neural network model, the method comprising:
[0006] Obtaining a microalgae sample, collecting Raman spectral data and determining target variable standard values for the microalgae sample to obtain original spectral data and variable standard values, wherein the target variables include microalgae cell dry weight and fucoxanthin yield;
[0007] Preprocessing the original spectral data and variable standard values to obtain a training set, a validation set, and a test set;
[0008] Constructing a component quantitative detection model, and obtaining an optimal parameter detection model by inputting the training set and the validation set into the component quantitative detection model for training and validation;
[0009] By inputting the test set into the optimal parameter detection model, the predicted value of the target variable is obtained.
[0010] Furthermore, the step of obtaining the microalgae sample includes: obtaining algae liquid at different culture time points under different light qualities and light intensities by culture, and obtaining the microalgae sample after centrifuging and collecting the algae liquid.
[0011] Furthermore, the step of collecting Raman spectral data includes: using an immersion Raman probe to collect Raman spectral data of the microalgae sample in real time under light-proof conditions.
[0012] Furthermore, the step of preprocessing the original spectral data and the variable standard value comprises:
[0013] The first spectral data is obtained by performing Savitzky-Golay smoothing on the original spectral data and then performing spectral data baseline correction based on an iterative discrete wavelet transform method;
[0014] The second spectral data is obtained by performing spectral interception on the first spectral data to remove redundancy and then performing standardization processing;
[0015] The second spectral data are labeled using the variable standard values, and then the labeled second spectral data are divided into a training set, a validation set, and a test set.
[0016] Furthermore, the component quantitative detection model is a one-dimensional convolutional neural network model, including an input layer, a convolution layer, a pooling layer, a Flatten layer, a Dropout layer, a fully connected layer and an output layer.
[0017] Furthermore, the step of inputting the training set and the validation set into the component quantitative detection model for training and validation includes:
[0018] The component quantitative detection model after the training set input parameters are initialized to obtain a first prediction result and complete a forward propagation;
[0019] Calculate the mean square error loss function through the first prediction result, obtain the gradient of the mean square error loss function with respect to the parameters of each layer, adaptively adjust the learning rate through the Adam optimizer, update the parameters of each layer, and complete a back propagation;
[0020] Repeat forward propagation and back propagation for multiple times to perform iterative training. After a specified number of iterations, the component quantitative detection model after iteration is verified through the verification set.
[0021] Furthermore, the step of verifying the iterated component quantitative detection model through the verification set includes:
[0022] The verification set is input into the iterated component quantitative detection model to obtain a second prediction result, and the determination coefficient and the root mean square error are calculated based on the second prediction result. When the determination coefficient and the root mean square error reach a preset threshold, the training is terminated to obtain a trained optimal parameter network model.
[0023] In a second aspect, the present application proposes a system for real-time in-situ analysis of biochemical components of microalgae by combining Raman spectroscopy detection and a neural network model. The system comprises a preprocessing module, a training module and an output module.
[0024] The preprocessing module is used to obtain microalgae samples, obtain variable standard values and original spectral data of target variables of the microalgae samples, and perform preprocessing to obtain training sets, validation sets, and test sets, wherein the variables include microalgae cell dry weight and fucoxanthin yield;
[0025] The training module is used to construct a component quantitative detection model, and the optimal parameter detection model is obtained by inputting the training set and the verification set into the component quantitative detection model for training and verification;
[0026] The output module is used to input the test set into the optimal parameter detection model to obtain the predicted value of the target variable.
[0027] The third aspect of the present application proposes an electronic device, which includes a processor and a memory, wherein a computer program is stored on the memory, and the processor is configured to implement the above-mentioned method of combining Raman spectroscopy detection and a neural network model for real-time in-situ analysis of biochemical components of microalgae when executing the computer program.
[0028] In a fourth aspect, the present application proposes a computer-readable storage medium, wherein the medium includes a stored computer program, and when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the above-mentioned method for real-time in-situ analysis of biochemical components of microalgae by combining Raman spectroscopy detection and a neural network model.
[0029] Compared with the prior art, the beneficial effects of this application are:
[0030] This application combines Raman spectroscopy technology with a one-dimensional convolutional neural network algorithm, and provides a method for real-time in-situ analysis of microalgae biochemical components by combining Raman spectroscopy detection and a neural network model, which establishes an effective relationship between Raman spectroscopy data and the standard value of the target variable, thereby enabling efficient, real-time, in-situ and non-invasive detection of microalgae biochemical components, obtaining key production parameters of microalgae cell populations, and more accurately predicting multiple parameters in practical applications. This application also discloses a system, electronic equipment and storage medium for real-time in-situ analysis of microalgae biochemical components by combining Raman spectroscopy detection and a neural network model, which also has the above-mentioned beneficial effects. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0032] Figure 1 A system flow chart of a method for real-time in-situ analysis of biochemical components of microalgae by combining Raman spectroscopy detection and a neural network model provided in the present application;
[0033] Figure 2 A schematic diagram of a method for real-time in-situ analysis of biochemical components of microalgae by combining Raman spectroscopy detection and a neural network model provided in the present application;
[0034] Figure 3 A schematic diagram of Raman spectral data of fucoxanthin standard mother solutions of different concentrations provided in the examples of the present application;
[0035] Figure 4 A schematic diagram of the original spectral data provided in the embodiments of the present application;
[0036] Figure 5 A schematic diagram for comparing the original spectral data and the first spectral data provided in an embodiment of the present application;
[0037] Figure 6 A schematic diagram of the structure of a component quantitative detection model provided in an embodiment of the present application;
[0038] Figure 7 The first variable standard value obtained by inputting the first training set into the partial least squares regression model for training is compared with the first variable predicted value provided in the embodiment of the present application;
[0039] Figure 8 The first variable standard value obtained by inputting the first validation set into the partial least squares regression model for training is compared with the first variable predicted value provided in the embodiment of the present application;
[0040] Fig. 9 Comparing the second variable standard value obtained by training the first training set into the partial least squares regression model provided in the embodiment of the present application with the second variable predicted value;
[0041] Fig.10 Comparing the second variable standard value obtained by training the first validation set into the partial least squares regression model provided in the embodiment of the present application with the second variable predicted value;
[0042] Fig.11 Comparison of the first variable standard value obtained by training the first training set input component quantitative detection model with the first variable predicted value provided in the embodiment of the present application;
[0043] Fig.12 Comparing the first variable standard value obtained by training the first validation set input component quantitative detection model with the first variable predicted value provided in the embodiment of the present application;
[0044] Fig.13 Comparison of the second variable standard value obtained by training the first training set input component quantitative detection model with the second variable predicted value provided in the embodiment of the present application;
[0045] Fig.14 Comparison of the second variable standard value obtained by training the first validation set input component quantitative detection model with the second variable predicted value provided in the embodiment of the present application;
[0046] Fig.15 Comparing the first variable standard value obtained by training the second training set input component quantitative detection model with the first variable predicted value provided in the embodiment of the present application;
[0047] Fig.16 Comparing the first variable standard value obtained by training the second validation set input component quantitative detection model with the first variable predicted value provided in the embodiment of the present application;
[0048] Fig.17 Comparison of the second variable standard value obtained by training the second training set input component quantitative detection model with the second variable predicted value provided in the embodiment of the present application;
[0049] Fig.18 Comparison of the second variable standard value obtained by training the second validation set input component quantitative detection model with the second variable predicted value provided in the embodiment of the present application;
[0050] Fig.19 The first variable standard value obtained by training the training set input component quantitative detection model provided in the embodiment of the present application is compared with the first variable predicted value;
[0051] Fig. 20 The first variable standard value obtained by training the validation set input component quantitative detection model provided in the embodiment of the present application is compared with the first variable predicted value;
[0052] Fig.21 The first variable standard value obtained by training the test set input component quantitative detection model provided in the embodiment of the present application is compared with the first variable predicted value;
[0053] Fig. 22 Comparison of the second variable standard value obtained by training the training set input component quantitative detection model provided in the embodiment of the present application with the second variable predicted value;
[0054] Fig.23 Comparison of the second variable standard value obtained by training the validation set input component quantitative detection model with the second variable predicted value provided in the embodiment of the present application;
[0055] Fig.24 Comparison of the second variable standard value obtained by training the test set input component quantitative detection model provided in the embodiment of the present application with the second variable predicted value;
[0056] Fig.25 This is a schematic diagram of the structure of the electronic device proposed in this application. DETAILED DESCRIPTION
[0057] The present application proposes a method for real-time in-situ analysis of microalgae biochemical components by combining Raman spectroscopy detection and a neural network model. In order to describe the present application in more detail, the technical solution of the present application is described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific implementation methods described herein are only used to explain the present application and are not used to limit the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.
[0058] In a first aspect, the present application proposes a method for real-time in-situ analysis of biochemical components of microalgae by combining Raman spectroscopy detection and a neural network model. The flowchart of the method is as follows: Figure 1 and Figure 2 As shown, it specifically includes the following steps:
[0059] S101, obtaining a microalgae sample, collecting Raman spectral data and determining target variable standard values for the microalgae sample to obtain original spectral data and variable standard values, wherein the target variables include microalgae cell dry weight and fucoxanthin yield;
[0060] Optionally, the microalgae are diatoms.
[0061] Wherein, the microalgae sample is a microalgae cell population, and the step of obtaining the microalgae sample comprises: obtaining algae liquid at different culture time points under different light qualities and light intensities by culture, and collecting the algae liquid by centrifugation to obtain the microalgae sample. The microalgae sample comprises a first microalgae sample and a second microalgae sample;
[0062] The step of collecting Raman spectroscopy data comprises: using an immersion Raman probe to collect Raman spectroscopy data of the microalgae sample in real time under light-proof conditions;
[0063] The step of determining the target variable standard value includes: determining the dry weight of microalgae cells of the microalgae sample to obtain a first variable standard value, and determining the fucoxanthin yield of the microalgae sample to obtain a second variable standard value.
[0064] S102, preprocessing the original spectral data and variable standard values to obtain a training set, a validation set, and a test set;
[0065] S103, constructing a component quantitative detection model, and obtaining an optimal parameter detection model by inputting the training set and the verification set into the component quantitative detection model for training and verification;
[0066] S104: Obtain a predicted value of a target variable by inputting the test set into the optimal parameter detection model.
[0067] The predicted value includes a first variable predicted value obtained by predicting the dry weight of microalgae cells of the microalgae sample and a second variable predicted value obtained by predicting the fucoxanthin yield of the microalgae sample.
[0068] In one embodiment, algae liquids at different culture time points under different light qualities and light intensities are obtained by laboratory autotrophic culture, and the algae liquids are centrifuged and collected to obtain the first microalgae sample. The specific steps of the laboratory autotrophic culture include:
[0069] S201, preparing a solid culture medium suitable for the growth of microalgae, and inoculating an appropriate amount of microalgae seeds into the sterilized solid culture medium for cultivation;
[0070] The solid culture medium is F / 2 enriched culture medium, which is a nutrient culture medium commonly used for culturing marine microalgae. Its specific formula is shown in the following table (Table 1):
[0071] Table 1 Culture medium formula
[0072]
[0073] S202, scrape the algae species growing on the solid culture medium, inoculate them into a 1 L triangular flask, place it in a shaking incubator for shaking culture, and set the culture conditions to 25 μmol / m 2 / s light intensity, white light quality and a rotation speed of 120 rpm;
[0074] S203. After 10 days of shaking culture, measure the OD of the algae solution in the triangular flask. 730 The required volume is calculated so that the algal solution has an initial OD of 0.2.730 The values were inoculated into 700 mL of liquid culture medium and placed in a column photobioreactor;
[0075] Among them, OD 730 The value refers to the initial optical density of the sample measured at a wavelength of 730 nm, which is used to estimate the concentration of algal cells and can be measured by a spectrophotometer and a cuvette. 730 The value is the OD value of the algae solution in the triangular flask. 730 Calculate the required volume The calculation formula is as follows:
[0076]
[0077] In this embodiment, the target initial OD 730 The value is 0.2, and the target culture volume is 700 mL.
[0078] The formula of the liquid culture medium is consistent with the specific formula of the solid culture medium (except agar).
[0079] S204, culturing in a column photobioreactor, setting the culture conditions to include: 25, 50, 100, 150, 300 μmol / m 2 / s light intensity, monochromatic light and mixed white light, and cultured with 1% CO2 / air for 7 days. The five light intensities and two light qualities were combined to form a total of 10 treatment groups;
[0080] S205. Collect algae liquid from different treatment groups every day from day 0 (the day of inoculation) to day 7, so as to obtain algae liquid at different culture time points under different light quality and light intensity conditions.
[0081] Specifically, in this embodiment, a total of 252 first microalgae samples under different light quality, light intensity and culture time conditions were obtained through laboratory autotrophic culture.
[0082] The training set includes a first training set and a second training set, and the validation set includes a first validation set and a second validation set. The first training set and the first validation set are obtained by collecting Raman spectral data and determining target variable standard values for the first microalgae sample, and preprocessing the obtained original spectral data and variable standard values, respectively, by dividing them into 70% and 30%.
[0083] In one embodiment, algae liquids at different culture time points under different light qualities and light intensities are obtained by culturing in a hybrid photobioreactor, and the algae liquids are centrifuged and collected to obtain the second microalgae sample, wherein the hybrid photobioreactor comprises:
[0084] A reactor body, wherein the reactor body is provided with a water tank for containing microalgae culture liquid; a light source device and a high-pressure washing and disinfection device are also assembled in the reactor body;
[0085] The light source device comprises a transparent sleeve and an LED light strip located in the transparent sleeve, and the transparent sleeves are arranged in parallel and spaced apart in the water tank;
[0086] The high-pressure flushing and disinfection device comprises a plurality of horizontal tubes, each of which is connected to a plurality of vertical tubes below, and each horizontal tube forms a comb-tooth structure with the vertical tubes connected thereto below; a plurality of liquid nozzles are arranged on the tube wall of each vertical tube from top to bottom; the lower end of the vertical tube is close to the bottom of the water tank; the plurality of horizontal tubes are connected to a total liquid inlet, and the total liquid inlet is connected to high-pressure clean water, disinfectant water or disinfectant steam; the projections of the transparent sleeve and the horizontal tubes on the bottom of the water tank are alternately spaced;
[0087] The reactor body is also assembled with a gas supply device, which includes at least one set of ventilation pipes, one end of which is connected to a gas supply source, and the ventilation pipes are laid at the bottom of the water tank, and a plurality of gas outlets are provided on the ventilation pipes;
[0088] It also includes a culture liquid temperature regulating device, which includes a plurality of heat exchangers. The heat exchangers are vertically installed in the water tank of the reactor body and are alternately arranged with the transparent sleeves and the horizontal tubes.
[0089] The specific steps of culturing by the hybrid photobioreactor include:
[0090] S301, filling the water tank of the hybrid photobioreactor with culture solution, and inoculating an appropriate amount of algae species into the culture solution;
[0091] S302, starting the hybrid photobioreactor for culturing, and presetting a plurality of treatment groups with different culturing conditions (light quality and light intensity);
[0092] Specifically, when natural light is insufficient or at night, the light source device is used to provide lighting conditions to the microalgae below the liquid surface, so that the algae cells in the algae liquid can receive light and promote their photosynthesis; the high-pressure flushing and disinfection device is used to clean and disinfect for the next round of microalgae cultivation, and when the culture liquid evaporates and becomes less, it can also be used to introduce the culture liquid, which can play a role in stirring the microalgae being cultivated; the air supply device is used to provide the algae liquid with CO2 (or introduce air) required for the growth of microalgae, and the floating air bubbles can also play a role in stirring the algae liquid, so that the culture medium is evenly dispersed, the algae liquid is prevented from sedimentation, and the algae cells in the water tank can grow and reproduce evenly; the temperature of the algae liquid in the water tank is effectively adjusted by the culture liquid temperature adjustment device.
[0093] S303. Collect algae liquid from different treatment groups every day from day 0 (the day of inoculation) to day 4, so as to obtain algae liquid at different culture time points under different light quality and light intensity conditions.
[0094] Specifically, in this embodiment, a total of 149 second microalgae samples under different light intensities and culture time conditions were obtained through the hybrid photobioreactor culture.
[0095] Among them, by collecting Raman spectral data and determining the standard value of the target variable for the second microalgae sample, and preprocessing the obtained original spectral data and variable standard values, the second training set and the second validation set are obtained by dividing them into 70% and 30% respectively.
[0096] In one embodiment, the Raman spectrum data is collected by a Raman spectrometer, which includes a laser, an immersion Raman probe, a darkroom, a spectrometer, and a corresponding control system and control software;
[0097] The laser is used to provide a stable laser beam to excite the molecules in the microalgae sample, thereby generating Raman scattering. In this embodiment, the excitation light wavelength of the laser beam is set to 830 nm;
[0098] The immersion Raman probe is used to directly insert into the sample for measurement, transmit the laser beam to the microalgae sample, and collect the scattered light back to the spectrometer;
[0099] The darkroom is used for light-proof measurement, and the microalgae sample and the immersion Raman probe are placed in the darkroom to avoid the influence of ambient light on Raman spectral data;
[0100] The spectrometer is connected to the immersion Raman probe via an optical fiber, and is used to separate the collected scattered light into Raman spectrum data of different wave numbers;
[0101] The control system corresponds to the control software and is used to control the operation and data acquisition of the Raman spectrometer, including setting parameters such as integration time, average number of scans, laser intensity, etc.
[0102] Specifically, the specific steps of collecting the Raman spectrum data by using a Raman spectrometer include:
[0103] S401. Before collecting the Raman spectrum data, the Raman spectrometer is calibrated once every three days by measuring the Raman spectrum of a cyclohexane standard sample to ensure detection accuracy;
[0104] Among them, cyclohexane is a commonly used calibration standard substance, and its Raman spectrum has clear characteristic peaks. By detecting the Raman spectrum peaks of cyclohexane standard samples, the wavenumber accuracy and spectral response of the Raman spectrometer can be verified.
[0105] Specifically, the cyclohexane standard sample was poured into a 10 mL quartz beaker and placed in a dark room, the cyclohexane standard sample submerged the immersion Raman probe, the control software preset the integration time to 200 ms, the average number of scans to 10 times, the laser intensity to 50%, and after removing the influence of ambient light or other background signals, the laser was turned on for Raman spectrum acquisition, the Raman spectrum peak of the cyclohexane standard sample was continuously detected, and the detection results were compared with the known cyclohexane peaks. The control software was calibrated and the instrument parameters were adjusted to ensure accurate instrument readings.
[0106] S402, before collecting the Raman spectrum data, predicting by measuring the Raman spectrum of the ethanol standard sample to ensure detection accuracy;
[0107] Among them, ethanol is also a commonly used calibration standard substance, and its Raman spectrum characteristic peak is clear and stable. Specifically, the ethanol standard sample is poured into a 10 mL quartz beaker and placed in a dark room. The ethanol standard sample is submerged in the immersion Raman probe. The control software presets the integration time to 300 ms, the average number of scans to 1, and the laser intensity to 50%. After removing the influence of ambient light or other background signals, the laser is turned on for Raman spectrum acquisition, and the Raman spectrum peak of the ethanol standard sample is continuously detected to ensure the accurate calibration of the Raman spectrometer and to ensure that the instrument can provide accurate and repeatable results during the experiment.
[0108] S403, after the microalgae sample is fully mixed with an oscillator, it is poured into a 10 mL quartz beaker and placed in a dark room, the microalgae sample is submerged in the immersion Raman probe, the control software is preset with an integration time of 2000 ms, an average number of scans of 10 times, a laser intensity of 50%, and after removing the influence of ambient light or other background signals, the laser is turned on to collect Raman spectra, and the Raman spectra data of the microalgae sample are repeatedly collected 10 times to obtain raw spectra data, and the raw spectra data are saved to a specified path.
[0109] The original spectral data include 2520 first original spectral data obtained by collecting the first microalgae sample 10 times, and 1490 second original spectral data obtained by collecting the second microalgae sample 10 times.
[0110] In one embodiment, the specific steps of determining the dry weight of microalgae cells of the microalgae sample include:
[0111] S501, taking 5 mL of the microalgae sample and filtering it through a pre-weighed (weight is DW0) 0.45 μm cellulose acetate membrane, and washing the microalgae cells on the cellulose acetate membrane with an equal volume of 0.5 mol / L NH4HCO3 to remove impurities;
[0112] S502, after the filtration is completed, the cellulose acetate membrane is placed in a 105°C oven to dry to constant weight (about 24 hours), then taken out and placed in a vacuum dryer to cool to room temperature and then weighed (weight is DW1);
[0113] Optionally, to facilitate filtration, remove excess NH4HCO3 and aid drying, the differential pressure of the vacuum pump is maintained at 35-55 mmHg.
[0114] S503, the calculation method of the dry weight dwt (g / L) of the microalgae cells of the microalgae sample is as follows:
[0115] dwt (g / L) = DW1- DW0
[0116] Two independent dry weight measurements are performed on the same sample to ensure the accuracy and repeatability of the measurement results, and the measurement results of each sample are recorded and averaged to obtain the first variable standard value of the microalgae sample.
[0117] In one embodiment, the fucoxanthin yield of the microalgae sample is determined by HPLC (High Performance Liquid Chromatography), and the specific steps include:
[0118] S601, weigh about 10 mg (DW) of the microalgae sample after freeze-drying. F ) was placed in a threaded tube, ground glass beads and 1 mL of methanol (HPLC grade) were added, and then the extract was obtained by multiple centrifugal extraction and filtration;
[0119] Specifically, S601 is performed in a light-proof environment, the grinding glass beads are 0.3 g, and the particle size is 0.4-0.6 mm, which are used to improve the homogenization and extraction efficiency of the freeze-dried microalgae sample through mechanical crushing, and then the 1 mL of HPLC grade methanol is used as a high-purity solvent to effectively dissolve the freeze-dried microalgae sample, and the solution obtained after the dissolution is subjected to cell crushing and centrifugation using a tissue homogenizer, and the supernatant is collected into a 10 mL brown volumetric flask, and the cell fragments are resuspended with 1 mL of methanol (HPLC grade) and centrifuged again, and the centrifugation operation is repeated until the precipitate becomes milky white, and finally the volume of the extracted supernatant is fixed to 10 mL, and after inversion and mixing, it is filtered with a 0.22 μm organic phase microporous filter membrane to remove tiny impurities to obtain an extract, which is placed in a chromatographic loading vial and prepared for liquid chromatography analysis.
[0120] S602, preparing standard working solutions of different concentrations using the fucoxanthin standard solution, collecting chromatograms of the standard working solutions of different concentrations using a high performance liquid chromatograph, and calculating the peak area of fucoxanthin, and establishing a fucoxanthin standard curve with the concentration of the fucoxanthin standard working solution as the abscissa and the fucoxanthin peak area in the chromatogram as the ordinate;
[0121] Specifically, first accurately weigh 1 mg of fucoxanthin standard solution and dissolve it in methanol (HPLC grade) solution, dilute to 5 mL, and prepare a fucoxanthin standard stock solution with a concentration of 200 μg / mL. Then, dilute step by step according to the following table (Table 2) to prepare a series of standard working solutions with known concentrations:
[0122] Table 2 Concentration preparation scheme of fucoxanthin standard working solution
[0123]
[0124] Secondly, start the high performance liquid chromatograph, which includes a pump, a column oven, a chromatographic column and a detector, etc. The mobile phase is pumped into the system by the pump to ensure that the mobile phase and the chromatographic column are balanced. In this embodiment, the detector adopts a photodiode array detector, and the chromatographic column adopts a reverse phase chromatographic column (C18 column) with a diameter of 4.6 mm, a length of 150 mm and a particle size of 5 μm. The high performance liquid chromatograph is set as follows: the composition of the mobile phase (for separating fucoxanthin) is: phase A is dichloromethane, phase B is methanol, phase C is acetonitrile, and phase D is water; the injection volume is 10 μL; the column temperature is 35°C; the detection wavelength is 450 nm (the characteristic absorption wavelength of fucoxanthin); standard working solutions of different concentrations are injected into the high performance liquid chromatograph in turn, and the gradient elution conditions are set to ensure that fucoxanthin is effectively separated and detected at the appropriate time point. The mobile phase elution gradient ratio is shown in the following table (Table 3), which shows the change of the mixing ratio of different solvents in the mobile phase composition over time:
[0125] Table 3 Mobile phase elution gradient ratio
[0126]
[0127] Record the chromatogram. In the obtained chromatogram, fucoxanthin will appear as an obvious peak. Use the analysis software of the HPLC to draw the fucoxanthin standard working curve with the concentration of the fucoxanthin standard working solution as the horizontal axis and the fucoxanthin peak area in the chromatogram as the vertical axis.
[0128] S603, R was obtained through the standard working curve of fucoxanthin 2>0.99, and the fucoxanthin content c μg / mL (the percentage of fucoxanthin in the dry weight of microalgae cells) of the extract is obtained by substituting the peak area of the extract into the linear regression equation, and then the second variable standard value is calculated by multiplying it by the first variable standard value (microalgae cell dry weight) corresponding to the microalgae sample.
[0129] The peak area of the extract is also obtained by collecting the chromatogram by the high performance liquid chromatograph. The final calculation formula of the fucoxanthin content (Fux%, the percentage of fucoxanthin in the dry weight of microalgae cells) of the microalgae sample is as follows:
[0130] Fux%=c×10 mL / 1000 / DW F ×100%
[0131] In one embodiment, a fucoxanthin standard solution (purity ≥ 95.0%, HPLC grade) is prepared into a fucoxanthin standard mother solution in 75% ethanol, and then a standard working solution of the required concentration of 0-200 ppm is prepared by a series of gradient dilutions, and stored in an environment of 4°C. At this time, a Raman spectrometer is used to collect spectral data of fucoxanthin standard working solutions of different concentrations, and the integration time is set to 200 ms, the average number of scans is set to 10 times, and the laser intensity is set to 50%, to obtain Raman spectral data of fucoxanthin standard working solutions of different concentrations.
[0132] The Raman spectral data of the fucoxanthin standard working solution with different concentrations are as follows: Figure 3 As shown, the Raman shift is 1160 cm -1 and 1531 cm -1 is the Raman characteristic peak of fucoxanthin, and the image of the original spectrum data is as follows Figure 4 As shown. Figure 3 As shown in the figure, with the increase of concentration, except for the Raman shift of 1160 cm -1 and 1531 cm -1 Except for the relative intensity increase of the characteristic peak at 1160 cm, the other characteristic peaks did not change significantly. It was thus identified that the main Raman characteristic peak of the fucoxanthin standard was located at 1160 cm -1 and 1531 cm -1 place; such as Figure 4 As shown in the Raman spectrum of the diatom algae liquid sample, a Raman shift of 1160 cm -1 and 1531 cm -1 The Raman characteristic peaks are consistent with the qualitative results of the standard.
[0133] In one embodiment, the step of preprocessing the variable standard value and the original spectral data comprises:
[0134] S701, performing Savitzky-Golay smoothing (ie, SG smoothing) on the original spectral data, and then performing spectral data baseline correction based on an iterative discrete wavelet transform method to obtain first spectral data;
[0135] The raw spectral data includes erroneous and unreliable information caused by the environment, sample, instrument, laser irradiation and random noise interference, so it is necessary to minimize redundant information and improve the signal-to-noise ratio of the spectrum through preprocessing to obtain the first spectral data;
[0136] The SG smoothing can effectively remove noise while retaining the main features of the spectral data. Specifically, the SG smoothing approximates the original signal by performing local polynomial fitting in a sliding window, and uses the fitting result to smooth the spectral data. The degree of smoothing is controlled by adjusting the window length and the polynomial fitting order, so as to adapt to spectral data of different types and complexities. The window length determines the number of data points used to calculate the weighted average of each data point, and the polynomial fitting order affects the degree of smoothing fitting.
[0137] Optionally, the SG smoothing sets the window length to 7 and the polynomial fitting order to 2.
[0138] The iterative discrete wavelet transform method is used to perform baseline correction on the spectral data processed by the SG smoothing algorithm, which can effectively remove the baseline drift in the spectrum, adaptively process linear and nonlinear baseline changes, and gradually make the spectral data flatter and more stable through multiple iterations. Specifically, the iterative discrete wavelet transform method uses the principle of discrete wavelet transform to first perform multi-scale wavelet decomposition on the spectral data after SG smoothing. At this time, the low-frequency component usually corresponds to the background or baseline of the spectrum, while the high-frequency component corresponds to more detailed signal changes. Then, the low-frequency component is approximated and eliminated or suppressed through multiple iterations, so that the effective spectral information is more prominent, which is helpful for extracting spectral features;
[0139] Optionally, the decomposition scale adopted by the iterative discrete wavelet transform method is 7.
[0140] The comparison diagram of the original spectrum data and the first spectrum data is as follows: Figure 5 As shown, Figure 5 The upper part is the original spectrum. Figure 5 The lower part is the first spectrum data after the original spectrum data is processed by Savitzky-Golay smoothing and de-iterative discrete wavelet transform method.
[0141] S702, performing spectrum interception on the first spectrum data to remove redundancy, and then performing standardization processing to obtain second spectrum data;
[0142] Wherein, the spectrum interception is used to retain effective spectrum information and remove redundant components, such as the Raman peak of the gem lens that may exist in the Raman probe;
[0143] The standardization process is used to scale the value range of the spectral data to a specific interval to ensure that the scale differences between different features do not have a negative impact. Specifically, the standardization process standardizes the spectral data after spectrum truncation by using the extreme value standardization (MinMaxScaler) method so that its value is linearly mapped to a given interval.
[0144] S703: label the second spectral data using the variable standard value, and then divide the labeled second spectral data into a training set, a validation set, and a test set.
[0145] The label production is used to generate corresponding annotation information to facilitate the training of the component quantitative detection model. High-quality labels are the basis for the successful training of the component quantitative detection model. If the label is inaccurate or erroneous, the model will learn wrong information during the training process, resulting in inaccurate prediction results. In this embodiment, each second spectral data corresponds to two labels, namely, the first variable standard value (microalgae cell dry weight) and the second variable standard value (fucoxanthin yield);
[0146] The training set is used for the subsequent training of the component quantitative detection model and the preliminary adjustment of parameters. The validation set is used to adjust the model's hyperparameters (such as learning rate, parameters, etc.) and evaluate the model's performance. The test set is used to finally evaluate the trained and validated model to obtain the predicted value of the target variable.
[0147] In one embodiment, Figure 6 As shown, the component quantitative detection model is a one-dimensional convolutional neural network model, including an input layer, a convolution layer, a pooling layer, a Flatten layer, a Dropout layer, a fully connected layer and an output layer;
[0148] The convolution layer includes a first convolution layer, a second convolution layer and a third convolution layer, the pooling layer includes a first pooling layer, a second pooling layer and a third pooling layer, and the component quantitative detection model gradually extracts multi-level abstract features of the input data from local features to more abstract overall features by stacking convolution layers and pooling layers;
[0149] The input layer receives input data, which contains 561 features, each of which has a dimension of 1, that is, each data is input into the model as a separate feature;
[0150] The first feature extraction is performed through the first convolutional layer to obtain a first feature map. The first convolutional layer includes 32 filters, each of which has a size of 3. The ReLU activation function is used, which helps to introduce nonlinear features and accelerate the convergence of the model. The He normal distribution is used for weight initialization, which helps to avoid the problem of gradient disappearance or explosion.
[0151] The spatial dimension of the first feature map is reduced by the first pooling layer to retain the most significant features, and the first maximum pooling layer adopts a maximum pooling operation with a window size of 2;
[0152] Performing a second feature extraction through the second convolutional layer to obtain a second feature map, the second convolutional layer includes 64 filters, the size of each filter is 3, and the ReLU activation function and He normal distribution initialization operation are also used;
[0153] The spatial dimension of the second feature map is reduced by the second pooling layer, and a maximum pooling operation with the same window size of 2 as the first pooling layer is adopted;
[0154] Performing a third feature extraction through the third convolutional layer to obtain a third feature map, the third convolutional layer includes 128 filters, the size of each filter is 3, and the ReLU activation function and He normal distribution initialization operation are also used;
[0155] Reducing the spatial dimension of the third feature map by the third pooling layer, and adopting a maximum pooling operation with the same window size of 2 as the first pooling layer and the second pooling layer;
[0156] The multi-dimensional third feature map is converted into a one-dimensional vector through the Flatten layer to prepare for the input of the fully connected layer;
[0157] Some neurons are randomly discarded through the Dropout layer to prevent overfitting, and the discard rate of the Dropout layer is 0.2;
[0158] The high-level features extracted from multiple convolutional layers and multiple pooling layers are learned through the fully connected layer to make the final output prediction. The fully connected layer includes 64 neurons and uses the ReLU activation function.
[0159] The regression task is performed through the output layer to generate a prediction result, and the output layer includes 1 neuron.
[0160] In one embodiment, the step of inputting the training set and the validation set into the component quantitative detection model for training and validation includes:
[0161] S801, inputting the component quantitative detection model after initializing the training set input parameters to obtain a first prediction result, completing a forward propagation;
[0162] The parameters include weights and biases. The initialization can alleviate the gradient vanishing problem and gradient exploding problem in deep networks. The gradient vanishing problem refers to the gradient value close to the input layer gradually becoming smaller or even close to zero as the number of network layers increases. The gradient exploding problem refers to the parameter gradient value becoming very large during the network model training process.
[0163] After the component quantitative detection model with the training set input parameters initialized, the training set is used as input data and passes through the input layer, the first convolution layer, the first pooling layer, the second convolution layer, the second pooling layer, the third convolution layer, the third pooling layer, the Flatten layer, the Dropout layer, the fully connected layer and the output layer in sequence.
[0164] S802, calculating a mean square error loss function using the first prediction result, obtaining a gradient of the mean square error loss function relative to the parameters of each layer, adaptively adjusting a learning rate using an Adam optimizer, updating the parameters of each layer, and completing a back propagation;
[0165] The forward propagation is used to calculate the predicted value of the input data through each layer structure of the network model, so as to compare it with the actual value and calculate the loss, and the back propagation is used to update the neural network weights and biases layer by layer by calculating the gradient of the loss function;
[0166] The weight refers to the coefficient connecting each neuron, which determines the weighted strength of the input signal when passing through the network layer. The bias refers to an additional parameter of each neuron, which enables the model to better fit the data. The gradient is the partial derivative of the loss function with respect to the parameters (weights and bias) in the neural network, which indicates the rate and direction of change of the loss function with the parameters. By calculating the gradient, we can know how to adjust the parameters to reduce the value of the loss function, thereby optimizing the model.
[0167] Specifically, the mean square error loss function is used to measure the difference between the model output and the target output to ensure that the model output is as close to the real data as possible. The calculation formula is as follows:
[0168]
[0169] in, Indicates the number of samples; represents the actual value (that is, the standard value of the variable in the label), Represents the model prediction value (that is, the first prediction result output by the model), Indicates error.
[0170] therefore, The smaller it is, the closer the model prediction value is to the actual value, and the better the model performs in the prediction task. During the training process, the model minimizes to optimize the parameters.
[0171] Among them, the learning rate is an important parameter that affects the efficiency of network model training. It is used to control the step size of each parameter update. Setting a too large learning rate will cause the loss function to fluctuate greatly, and the parameters will jump on both sides of the optimal value, making it difficult for the network model to converge during training. Setting a too small learning rate will easily lead to overfitting during network model training, and the parameter optimization speed will be reduced.
[0172] In one embodiment, the Adam optimizer adjusts the learning rate by calculating the first-order and second-order moment estimates of the gradient of each parameter, thereby accelerating convergence and improving model performance. Specifically, in S801, the first-order moment estimate is also initialized. and the second moment estimate , preset learning rate α and first-order moment estimation decay rate and the second-order moment estimate decay rate ;
[0173] In S802, the gradient of the loss function with respect to the parameter θ is calculated. , and based on the gradient and the decay rate, the first-order moment estimate and the second-order moment estimate are updated, and the calculation formula is as follows:
[0174] because and There is a deviation at the beginning, and deviation correction is required. The calculation formula is as follows:
[0175] Then use the modified first-order moment estimate and second-order moment estimate to update the parameters. The calculation formula is as follows:
[0176] Wherein, t represents the number of iterations, , , and They represent the first-order moment estimate, the second-order moment estimate, the gradient of the loss function relative to the parameters, and the parameters at this time at the t-th iteration, respectively. Represents a numerical stability term (used to prevent division by zero errors when updating parameters and ensure the stability of numerical calculations).
[0177] The adaptive moment estimation algorithm shows strong robustness in practice due to its adaptability and numerical stability design, can handle unstable gradients and noisy objective functions, and is suitable for various types of neural networks and machine learning tasks.
[0178] S803, repeating forward propagation and back propagation for multiple times to perform iterative training, and after a specified number of iterations, verifying the component quantitative detection model after iteration through the verification set;
[0179] In one embodiment, the prescribed number of times is 100 rounds and the batch size is 32.
[0180] S804, inputting the verification set into the iterated component quantitative detection model to obtain a second prediction result;
[0181] After the component quantitative detection model with the validation set input parameters initialized, the training set is used as input data and passes through the input layer, the first convolution layer, the first pooling layer, the second convolution layer, the second pooling layer, the third convolution layer, the third pooling layer, the Flatten layer, the Dropout layer, the fully connected layer and the output layer in sequence.
[0182] S805. Calculate the determination coefficient and the root mean square error through the second prediction result. When the determination coefficient and the root mean square error reach a preset threshold, the training is terminated to obtain a trained optimal parameter network model.
[0183] The coefficient of determination R 2 (R-squared) is used to measure the degree of fit of the model to the data. The value range is between 0 and 1. The closer it is to 1, the better the model fits the data. The calculation formula is as follows:
[0184] in, Indicates the number of samples; represents the actual value (that is, the standard value of the variable in the label), Represents the model prediction value (that is, the second prediction result output by the model), is the average of the actual values.
[0185] The root mean square error RMSE is used to quantify the prediction error of the model, and the calculation formula is as follows:
[0186] in, Indicates the number of samples; represents the actual value (that is, the standard value of the variable in the label), Represents the model prediction value (that is, the second prediction result output by the model).
[0187] therefore, The closer it is to 1, The smaller the value, the closer the model prediction value is to the actual value, and the better the model performs in the prediction task. As close to 1 as possible and minimize to optimize the parameters.
[0188] Optionally, the validation set monitors and dynamically adjusts the training process through EarlyStopping (early stopping method) and / or ReduceLROnPlateau (learning rate decay) callback functions.
[0189] Specifically, the EarlyStopping callback function is used to monitor the performance indicators (i.e., the coefficient of determination and the root mean square error) on the validation set, and terminate the training early when the performance indicators no longer improve significantly within the preset rounds, so as to prevent overfitting and save computing resources;
[0190] The ReduceLROnPlateau callback function is used to monitor the performance indicators (i.e., the coefficient of determination and the root mean square error) on the validation set, and reduce the learning rate when the performance indicators no longer improve significantly within the preset rounds, so as to help the model adjust parameters more carefully and converge to a better solution more stably in the later stages of training;
[0191] In one embodiment, the first training set and the first validation set are input into a partial least squares regression (PLSR) model for training and validation. Figure 7 , Figure 8 Comparison of the first variable standard value and the first variable predicted value for the first training set and the first validation set, respectively (in g / L); Fig. 9 , Fig.10 Comparison of the standard value of the second variable with the predicted value of the second variable for the first training set and the first validation set, respectively (in mg / L).
[0192] The partial least squares regression model is a classic linear chemometric regression model that avoids multicollinearity of data by projecting spectral data onto latent variables that explain the variance within the spectral data. It is applicable to many chemometric data sets, but these data sets often consist of a large number of variables and relatively few samples.
[0193] according to Figure 7 , Figure 8 , Fig. 9 and Fig.10 The results of quantitative analysis of the dry weight of microalgae cells showed that the root mean square error RMSE of the training set was 0.2106 g / L, and the linear regression fit R 2 The RMSE of the validation set is 0.9362, and the RMSE of the validation set is 0.2034 g / L. 2 The quantitative analysis results of fucoxanthin production showed that the RMSE of the training set was 1.6310 mg / L, R 2 is 0.9552; the RMSE of the validation set is 1.7239 mg / L, R 2It is 0.9510.
[0194] In one embodiment, the first training set and the first validation set are input into a component quantitative detection model for training and validation. Fig.11 , Fig.12 Comparison of the first variable standard value and the first variable predicted value for the first training set and the first validation set, respectively (in g / L); Fig.13 , Fig.14 Comparison of the standard value of the second variable with the predicted value of the second variable for the first training set and the first validation set, respectively (in mg / L).
[0195] according to Fig.11 , Fig.12 , Fig.13 and Fig.14 The quantitative analysis results of the dry weight of microalgae cells showed that the root mean square error RMSE of the first training set was 0.089 g / L, which was 57.7% lower than that of the PLSR model, and the linear regression fit R 2 The RMSE of the first validation set is 0.1279 g / L, which is 37.2% lower than that of the PLSR model. 2 The quantitative analysis results of fucoxanthin production showed that the RMSE of the first training set was 0.6905 mg / L, which was 57.6% lower than that of PLSR. 2 The RMSE of the first validation set is 1.1027 mg / L, which is 43.7% lower than that of PLSR. 2 It is 0.9713. Obviously, the component quantitative detection model is more accurate than the PLSR model.
[0196] In one embodiment, the second training set and the second validation set are input into the component quantitative detection model for training and validation. Fig.15 , Fig.16 Comparison of the first variable standard value and the first variable predicted value for the second training set and the second validation set, respectively (in g / L); Fig.17 , Fig.18 Comparison of the standard value of the second variable with the predicted value of the second variable for the second training set and the second validation set, respectively (in mg / L).
[0197] according to Fig.15 , Fig.16 , Fig.17 and Fig.18 The results of quantitative analysis of the dry weight of microalgae cells showed that the RMSE of the second training set was 0.089 g / L, and R 2 The RMSE of the second validation set is 0.9805; the RMSE of the second validation set is 0.0441 g / L, R 2The quantitative analysis results of fucoxanthin production showed that the RMSE of the second training set was 0.2878 mg / L, R 2 The RMSE of the second validation set is 0.9767; the RMSE of the second validation set is 0.4937 mg / L, R 2 It is 0.9164.
[0198] In one embodiment, 33 more microalgae samples (including 10 first microalgae samples and 23 second microalgae samples) are collected, and the corresponding pre-processed raw spectral data and variable standard values are used as a test set. The training set, validation set and test set are input into the component quantitative detection model for training, validation and testing. Fig.19 , Fig. 20 , Fig.21 Comparison of the first variable standard value and the first variable predicted value of the training set, validation set and test set, respectively (in g / L); Fig. 22 , Fig.23 , Fig.24 Comparison of the second variable standard value and the second variable predicted value for the training set, validation set and test set, respectively (in mg / L).
[0199] according to Fig.19 , Fig. 20 , Fig.21 , Fig. 22 , Fig.23 and Fig.24 The quantitative analysis results of the dry weight of microalgae cells showed that the RMSE of the training set was 0.0997 g / L and R 2 The validation set RMSE is 0.9812; the validation set RMSE is 0.1214 g / L, R 2 The RMSE of the test set is 0.9716; the RMSE of the test set is 0.0484 g / L, R 2 The quantitative analysis results of fucoxanthin production showed that the RMSE of the training set was 0.9991 mg / L, and R 2 The validation set RMSE is 1.2370 mg / L, R 2 The RMSE of the test set is 0.9642; the RMSE of the test set is 0.6868 mg / L, R 2 It is 0.9718.
[0200] In a second aspect, the present application proposes a system for real-time in-situ analysis of biochemical components of microalgae by combining Raman spectroscopy detection and a neural network model. The system comprises a preprocessing module, a training module and an output module.
[0201] The preprocessing module is used to obtain microalgae samples, obtain variable standard values and original spectral data of target variables of the microalgae samples, and perform preprocessing to obtain training sets, validation sets, and test sets, wherein the variables include microalgae cell dry weight and fucoxanthin yield;
[0202] The training module is used to construct a component quantitative detection model, and the optimal parameter detection model is obtained by inputting the training set and the verification set into the component quantitative detection model for training and verification;
[0203] The output module is used to input the test set into the optimal parameter detection model to obtain the predicted value of the target variable.
[0204] In one embodiment, the component quantitative detection model is a one-dimensional convolutional neural network, including an input layer, a convolution layer, a pooling layer, a dropout layer, a flatten layer, a fully connected layer and an output layer;
[0205] The convolution layer includes a first convolution layer, a second convolution layer and a third convolution layer, and the pooling layer includes a first pooling layer, a second pooling layer and a third pooling layer;
[0206] The input layer is used to receive input data, including 561 features, and the dimension of each feature is 1;
[0207] The first convolutional layer is used to perform the first feature extraction to obtain the first feature map. The first convolutional layer includes 32 filters, each of which has a size of 3. The ReLU activation function is used, which helps to introduce nonlinear features and accelerate the convergence of the model. The He normal distribution is used for weight initialization, which helps to avoid the problem of gradient disappearance or explosion.
[0208] The first pooling layer is used to reduce the spatial dimension of the first feature map and retain the most significant features, and the first maximum pooling layer adopts a maximum pooling operation with a window size of 2;
[0209] The second convolutional layer is used to perform a second feature extraction to obtain a second feature map. The second convolutional layer includes 64 filters, each of which has a size of 3 and also uses a ReLU activation function and a He normal distribution initialization operation;
[0210] The second pooling layer is used to reduce the spatial dimension of the second feature map, and adopts the maximum pooling operation of the same window size 2 as the first pooling layer;
[0211] The third convolutional layer is used to perform a third feature extraction to obtain a third feature map. The third convolutional layer includes 128 filters, each of which has a size of 3, and also uses a ReLU activation function and a He normal distribution initialization operation;
[0212] The third pooling layer is used to reduce the spatial dimension of the third feature map, and adopts a maximum pooling operation of the same window size 2 as the first pooling layer and the second pooling layer;
[0213] The Flatten layer is used to convert the multi-dimensional third feature map into a one-dimensional vector to prepare for the input of the fully connected layer;
[0214] The Dropout layer is used to randomly discard some neurons to prevent overfitting, and the discard rate of the Dropout layer is 0.2;
[0215] The fully connected layer is used to learn high-level features extracted from multiple convolutional layers and multiple pooling layers to perform final output prediction. The fully connected layer includes 64 neurons and uses a ReLU activation function.
[0216] The output layer is used to perform regression tasks and generate prediction results, and the output layer includes 1 neuron;
[0217] The component quantitative detection model is connected according to the structure of input layer, first convolution layer, first pooling layer, second convolution layer, second pooling layer, third convolution layer, third pooling layer, Flatten layer, Dropout layer, fully connected layer and output layer.
[0218] The third aspect of the present application provides an electronic device, such as Fig.25 As shown, the device includes a processor and a memory, the memory stores a computer program, and the processor is configured to implement the above-mentioned method for real-time in-situ analysis of microalgae biochemical components by combining Raman spectroscopy detection and a neural network model when executing the computer program.
[0219] Optionally, the computer program may be divided into one or more modules / units (such as computer program 1, computer program 2, etc.), and the one or more modules / units are stored in the memory and executed by the processor. The one or more modules / units may be a series of computer program instruction segments that can perform specific functions, and the instruction segments are used to describe the execution process of the computer program in the computer device.
[0220] Optionally, the electronic device may include one or more processors, which may be a central processing unit (CPU), a microprocessor (MPU), or other general-purpose processors, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices (PLD), discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or the processor may be any conventional processor. The processor is the control center of the computer device, and various parts of the computer device are connected using various interfaces and lines.
[0221] Optionally, the electronic device may include one or more memories, the memory mainly including a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function, etc., and the data storage area may store relevant data, etc. In addition, the memory may be a high-speed random access memory, or a non-volatile memory, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card and a flash card (Flash Card), etc., or other volatile solid-state storage devices. In some embodiments, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to a computer terminal via a network. Embodiments of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0222] Optionally, the electronic device may also communicate with one or more external devices (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device, and / or communicate with any device that enables the electronic device to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface. Furthermore, the electronic device may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter. Other hardware and / or software modules may be used in conjunction with the electronic device, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.
[0223] In a fourth aspect, the present application proposes a computer-readable storage medium, wherein the medium includes a stored computer program, and when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the above-mentioned method for real-time in-situ analysis of biochemical components of microalgae by combining Raman spectroscopy detection and a neural network model.
[0224] Optionally, the computer readable storage medium may be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection with one or more conductors, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.
[0225] Program code for performing the operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages such as Java, C++, etc., and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., via the Internet using an Internet service provider).
[0226] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Those skilled in the art should be aware that there may be many examples based on the basic method principles provided by the present application in combination with actual conditions, and all of them should be within the scope of protection of the present application without sufficient creative work.
[0227] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms are not necessarily for the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.
[0228] It should also be noted that, in this specification, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.
Claims
1. A real-time in-situ detection method for biochemical components of microalgae based on Raman spectroscopy, characterized in that: The method comprises: algae liquids at different culture time points under different light qualities and light intensities are obtained by culturing, and the algae liquids are centrifuged and collected to obtain microalgae samples; Performing Raman spectral data collection and target variable standard value determination on the microalgae sample to obtain original spectral data and variable standard values, wherein the target variables include microalgae cell dry weight and fucoxanthin yield; wherein the Raman spectral data collection includes exciting the microalgae sample with an excitation light wavelength of 830 nm; Preprocessing the original spectral data and variable standard values to obtain a training set, a validation set, and a test set; Constructing a component quantitative detection model, and obtaining an optimal parameter detection model by inputting the training set and the verification set into the component quantitative detection model for training and verification; the component quantitative detection model is a one-dimensional convolutional neural network model; By inputting the test set into the optimal parameter detection model, the predicted value of the target variable is obtained.
2. The real-time in-situ detection method of microalgae biochemical components based on Raman spectroscopy according to claim 1, characterized in that: The step of collecting Raman spectrum data includes: using an immersion Raman probe to collect Raman spectrum data of the microalgae sample in real time under light-proof conditions.
3. The real-time in-situ detection method of microalgae biochemical components based on Raman spectroscopy according to claim 1, characterized in that: The step of preprocessing the original spectral data and the variable standard value comprises: The first spectral data is obtained by performing Savitzky-Golay smoothing on the original spectral data and then performing spectral data baseline correction based on an iterative discrete wavelet transform method; The second spectral data is obtained by performing spectral interception on the first spectral data to remove redundancy and then performing standardization processing; The second spectral data are labeled using the variable standard values, and then the labeled second spectral data are divided into a training set, a validation set, and a test set.
4. The real-time in-situ detection method of microalgae biochemical components based on Raman spectroscopy according to claim 1, characterized in that: The component quantitative detection model includes an input layer, a convolution layer, a pooling layer, a Flatten layer, a Dropout layer, a fully connected layer and an output layer.
5. The real-time in-situ detection method of microalgae biochemical components based on Raman spectroscopy according to claim 1, characterized in that: The step of inputting the training set and the validation set into the component quantitative detection model for training and validation comprises: The component quantitative detection model after the training set input parameters are initialized to obtain a first prediction result and complete a forward propagation; Calculate the mean square error loss function through the first prediction result, obtain the gradient of the mean square error loss function with respect to the parameters of each layer, adaptively adjust the learning rate through the Adam optimizer, update the parameters of each layer, and complete a back propagation; Repeat forward propagation and back propagation for multiple times to perform iterative training. After a specified number of iterations, the component quantitative detection model after iteration is verified through the verification set.
6. The real-time in-situ detection method of microalgae biochemical components based on Raman spectroscopy according to claim 5, characterized in that: The step of verifying the iterated component quantitative detection model through the verification set includes: The verification set is input into the iterated component quantitative detection model to obtain a second prediction result, and the determination coefficient and the root mean square error are calculated based on the second prediction result. When the determination coefficient and the root mean square error reach a preset threshold, the training is terminated to obtain a trained optimal parameter network model.
7. A real-time in-situ detection system for microalgae biochemical components based on Raman spectroscopy, characterized in that: The system includes a preprocessing module, a training module and an output module. The preprocessing module is used to obtain microalgae samples, obtain the variable standard value and original spectral data of the target variable of the microalgae sample, and perform preprocessing to obtain a training set, a validation set, and a test set, wherein the variables include microalgae cell dry weight and fucoxanthin yield; wherein algae liquid at different culture time points under different light qualities and light intensities is obtained by culture, and the algae liquid is centrifuged and collected to obtain the microalgae sample; the collection of the original spectral data includes exciting the microalgae sample by an excitation light wavelength of 830 nm; The training module is used to construct a component quantitative detection model, and the optimal parameter detection model is obtained by inputting the training set and the verification set into the component quantitative detection model for training and verification; wherein the component quantitative detection model is a one-dimensional convolutional neural network model; The output module is used to input the test set into the optimal parameter detection model to obtain the predicted value of the target variable.
8. An electronic device, characterized in that: The device comprises a processor and a memory, wherein a computer program is stored in the memory, and the processor is configured to implement the method according to any one of claims 1 to 6 when executing the computer program.
9. A computer-readable storage medium, characterized in that: The medium includes a stored computer program, and when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to any one of claims 1 to 6.
