CN114239399B - Spectral data enhancement method based on conditional variation self-coding - Google Patents
Spectral data enhancement method based on conditional variation self-coding Download PDFInfo
- Publication number
- CN114239399B CN114239399B CN202111546557.1A CN202111546557A CN114239399B CN 114239399 B CN114239399 B CN 114239399B CN 202111546557 A CN202111546557 A CN 202111546557A CN 114239399 B CN114239399 B CN 114239399B
- Authority
- CN
- China
- Prior art keywords
- spectrum
- semi
- supervised
- ladder network
- virtual
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention discloses a spectral data enhancement method based on conditional variation self-coding, which mainly comprises the following four steps: step one: preprocessing the collected historical data; step two: training a condition-variable automatic encoder to generate virtual samples; step three: training a semi-supervised ladder network to construct a calibration model; step four: and (3) detecting the component content on line by using a calibration model constructed by a semi-supervised ladder network. The condition variation automatic encoder aims at generating a virtual spectrum with the same distribution as the component concentration so as to strengthen a training set and facilitate the development of a calibration model, and a regression learning model based on a semi-supervised ladder network is used for modeling by using the generated virtual spectrum; the method combines the generated virtual unlabeled spectrum with the real labeled sample, can ensure that the generated virtual spectrum and the actual labeled spectrum have the same distribution, and ensures the effectiveness of semi-supervised learning.
Description
Technical Field
The invention relates to a spectrum data enhancement method, in particular to a spectrum data enhancement method based on conditional variation self-coding.
Background
In recent years, near infrared spectrum technology is easy to operate on line, has strong anti-interference capability and high measurement precision, and is therefore increasingly popular in-situ measurement of biological and chemical processes. A key issue in component measurement using spectroscopic techniques is the development of calibration models. Spectral calibration modeling can be seen as a regression task. Due to the complexity of the production process, such as environmental disturbances, operating conditions effects, etc., the production process often exhibits complex nonlinearities, which result in a highly nonlinear relationship between spectral variables and component content. The traditional machine learning method is difficult to effectively extract the characteristics of the data, so that the prediction effect of the model is not ideal.
In recent years, deep learning methods typified by neural networks have been used to model complex and nonlinear process data. However, for a deep learning based regression task, a large amount of sample data and its corresponding labels are required, however, it is difficult for the process industry to acquire a large amount of data, especially for a calibration model task. Marking unlabeled data requires expert manipulation, laboratory specific equipment, and analysis in time, which is laborious and time consuming.
One approach to the problem of lack of labeled samples is to use unlabeled data to enhance the parametric information of the model. Semi-supervised methods can effectively utilize unlabeled samples to obtain more accurate data distribution. However, semi-supervised learning is based on the assumption that both marked and unmarked samples are generated from the same distribution. Given the large number of unlabeled samples and the interference of environmental conditions, it is difficult to ensure that all unlabeled samples come from the same distribution as the labeled samples. In this case, the introduction of unlabeled samples can compromise the accuracy of the model. Another approach is to extend the data set so that data enhancement can be used to enhance the model parameters. The data enhancement method first synthesizes virtual samples for enhancing the data set, and then uses the expanded data set in combination with the original data to construct a regression quantity. One representative data enhancement method is a generative antagonism network, which has been applied to classification problems. But the generated countermeasure network model is difficult to train and is prone to collapse and non-convergence, which is unsuitable for practical applications, especially for spectral calibration. Another important class of representative data enhancement methods is the variational auto-encoders, which approximate the original data distribution in the latent variable space. However, the data enhancement method based on the variation automatic encoder does not consider the required target information, which makes such a data enhancement method lack directionality, and thus some scholars propose a data enhancement method based on the conditional variation automatic encoder. The proposed data enhancement method based on a conditional variable automatic encoder is still only effective for classification tasks. The regression task of being continuous for the tag values is not yet efficient. The conditional variation automatic encoder can ensure that the generated samples and the original data belong to the same distribution, but the same tag value can generate different samples, and if the generated samples and tags are directly used for data modeling, the enhanced data cannot be guaranteed to be beneficial to model construction. Therefore, the invention provides a spectrum calibration modeling method combining a condition variation automatic encoder and a semi-supervised ladder network.
Disclosure of Invention
Aiming at the problem of insufficient marking samples of a spectrum calibration model, the invention provides a spectrum data enhancement method for spectrum calibration modeling based on conditional variation self-coding.
The invention is realized by the following technical scheme: a method of spectral data enhancement based on conditional variation self-encoding, comprising the steps of:
Step one: preprocessing collected historical data
Measuring the component concentration of the product by using an instrument to obtain a labeled training set comprising a labeled spectrum X L＝[x1;x2;xi;…;xL and a corresponding component concentration Y L, drawing a trend chart of the obtained component concentration of the product, and deleting the spectrum and the component content which do not accord with the process trend from the training set; then, the clean training sets X L and Y L are subjected to standardization treatment so as to weaken the influence of external interference on the spectrum;
Step two: establishing a conditional variation self-encoder based on an encoder q (z|x, Y) and a decoder p (x|z, Y), setting a virtual component content Y U in a modeling set to generate a virtual spectrum X U, wherein X and Y are a spectrum and a component concentration of a training set respectively, p represents a probability distribution function, and z represents a latent variable;
generating a similar sample of representative marker samples, written as p (x|y)
p(x|y)＝∫p(z)p(x|y,z)dz
The probability model, that is, maximizing the collected data, is built as:
wherein logp (x|y) is further decomposed into:
Wherein L ELBO is the variation lower bound;
Since the value of the non-negativity of the KL divergence is greater than or equal to 0, there are
logp(x|y)≥LELBO
Wherein L ELBO is further decomposed into:
LELBO＝-KL(q(z|x,y)||p(z|y))+∫q(z|x,y)logp(x|z,y)dz
step three: training a semi-supervised ladder network to construct a calibration model;
Step four: and inputting the online spectrum sample into a semi-supervised ladder network spectrum calibration model guided by an established condition variation self-encoder, and detecting the component content online in real time.
Preferably, the standardized processing formula of the training sets X L and Y L in the first step is as follows:
wherein: x i,std represents the absorbance of the ith spectral data subjected to normalization treatment; x i represents the absorbance of the ith spectral data wave; m represents the average value of absorbance of the spectrum data with the label; s represents the standard deviation of absorbance of the spectral data with the label.
Preferably, the virtual component content Y U in the second step is within the range of Y L, the step length is set, and the number of generated virtual spectrums is more than 10 times of the number of tagged data.
Preferably, in the third step, the virtual sample X U generated by using the condition variation self-encoder is input to the non-label part of the semi-supervised ladder network, the real marked samples X L and Y L are input to the labeled part of the semi-supervised ladder network, and semi-supervised learning is performed by adopting the semi-supervised ladder network, so as to obtain the spectrum calibration model of the semi-supervised ladder network guided by the condition variation self-encoder.
Compared with the prior art, the invention has the following advantages:
1. Compared with the method for establishing a semi-supervised calibration model by using real non-tag data in the process, the method can ensure that the generated virtual spectrum and the tagged spectrum belong to the same distribution, and avoid the problem that the accuracy of the model is damaged due to the fact that the real non-tag deviates from the real distribution caused by external interference;
2. The condition variation automatic encoder aims at generating a virtual spectrum with the same distribution as the component concentration so as to strengthen a training set and facilitate development of a calibration model, and a regression learning model based on a semi-supervised ladder network is used for modeling by using the generated virtual spectrum; the method combines the generated virtual unlabeled spectrum with the real labeled sample, can ensure that the generated virtual spectrum and the actual labeled spectrum have the same distribution, and ensures the effectiveness of semi-supervised learning.
Drawings
The invention is further described below with reference to the accompanying drawings.
FIG. 1 is a flow chart of a method of spectral data enhancement for spectral calibration modeling based on conditional variation self-encoding;
fig. 2 is a method of spectral data generation from an encoder based on conditional variations.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. It will be apparent that the described embodiments are only some, but not all, embodiments of the invention. Thus, the following detailed description of the embodiments of the invention, as presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be made by a person skilled in the art without making any inventive effort, are intended to be within the scope of the present invention.
Example 1
A method of spectral data enhancement based on conditional variation self-encoding, comprising the steps of:
Step one: preprocessing the collected historical data;
when the spectra are measured in real time, the process is sampled at regular intervals, and the concentration of the product components is measured by using a specific instrument, so that a labeled training set comprising the labeled spectra X L＝[x1;x2;xi;…;xL and the corresponding component concentrations YL is obtained. Since manual labeling may create operational errors, the labeled samples may compromise the accuracy of the model, and thus must be excluded from the training set. When the sample is removed, drawing a trend graph of the obtained component concentration of the product, and deleting the spectrum and the component content which do not accord with the process trend from the training set; then, the clean training sets X L and Y L are subjected to standardization treatment so as to weaken the influence of external interference on the spectrum;
wherein: x i,std represents the absorbance of the ith spectral data subjected to normalization treatment; x i represents the absorbance of the ith spectral data wave; m represents the average value of absorbance of the spectrum data with the label; s represents the standard deviation of absorbance of the spectrum data with the label;
In the embodiment, a spectrum is obtained every 1 minute, sampling is carried out every half hour, and a sampling instrument is a near infrared spectrum;
Step two: establishing a conditional variation self-encoder based on an encoder q (z|x, Y) and a decoder p (x|z, Y), setting a virtual component content Y U in a modeling set to generate a virtual spectrum X U, wherein X and Y are a spectrum and a component concentration of a training set respectively, p represents a probability distribution function, and z represents a latent variable;
Training a condition variation automatic encoder to generate a virtual spectrum sample X U is a key step of establishing a spectrum semi-supervised calibration model;
generating a similar sample of representative marker samples, written as p (x|y)
p(x|y)＝∫p(z)p(x|y,z)dz
The probability model, that is, maximizing the collected data, is built as:
wherein logp (x|y) is further decomposed into:
Wherein L ELBO is the variation lower bound;
Since the value of the non-negativity of the KL divergence is greater than or equal to 0, there are
logp(x|y)≥LELBO
Wherein L ELBO is further decomposed into:
LELBO＝-KL(q(z|x,y)||p(z|y))+∫q(z|x,y)logp(x|z,y)dz
Note that the virtual component content Y U here is within the range of Y L, the step size is set, and the number of generated virtual spectrums is 10 times or more the number of tagged data;
step three: training a semi-supervised ladder network to construct a calibration model;
In the step, a virtual sample X U generated by using a condition variation self-encoder is input to an unlabeled part of a semi-supervised ladder network, real marked samples X L and Y L are input to a labeled part of the semi-supervised ladder network, semi-supervised learning is performed by adopting the semi-supervised ladder network, and a spectrum calibration model of the semi-supervised ladder network guided by the condition variation self-encoder is obtained. The specific experimental reference "Semi-SupervisedLearning-BasedCalibrationModelBuilding ofNIR Spectroscopy for In Situ Measurement ofBiochemical Processes Under Insufficiently and Inaccurately Labeled Samples,"IEEE Transactions on Instrumentation andMeasurement,vol.70,pp.1-12,2021,Artno.2509912, to a semi-supervised ladder network is not a protection of the present invention herein.
Step four: the component content is detected on line by using a calibration model constructed by a semi-supervision ladder network;
and inputting the online spectrum sample into a semi-supervised ladder network spectrum calibration model guided by an established condition variation self-encoder, and detecting the component content online in real time.
Example 2
To verify the effectiveness of the present invention, experimental verification of converting glucose to ethanol at a concentration of 120g/L in a5 liter fermentor was performed.
The pH of the fermenter was adjusted by NaOH solution, the temperature was measured by Pt100, and the temperature was adjusted by a heating device and a recirculating chiller. The stirrer speed was set at 200rpm to maintain a uniform liquid phase. Each batch run time of the process was 12-15 hours. Near infrared spectrometers with immersed transmission probes detect fermentation processes in real time using a sampling period of one minute. At the same time, 5mL of liquid was taken out of the fermenter every half an hour, and the ethanol concentration was measured off-line using a gas chromatograph. In real-time measurements, there are 936 spectral variables per collected near infrared spectrum, ranging from 4,790 to 12,000cm -1.
Eight batches of glucose fermentation experiments were performed, seven of which were used for calibration modeling of near infrared spectra, and the eighth for model verification.
To illustrate the effectiveness of the method, the method is compared with the existing methods including eight methods of partial least squares, least squares support vector machines, denoising self-encoders, denoising source separation, supervisory ladder network, semi-supervisory denoising self-encoders, semi-supervisory denoising source separation and semi-supervisory ladder network. The optimal model parameters are obtained through cross validation by the partial least square and least square support vector machine, the hidden layer of the deep learning method is set to be 5, the neuron number of each layer is 1000,500,300,200,100, the learning rate is 0.003, the training batch size is 300, and the noise excitation level is 0.12. The first eight methods are specific experimental procedures and experimental results, see reference "Semi-Supervised Learning-Based Calibration Model Building of NIR Spectroscopy for In Situ Measurement of Biochemical Processes Under Insufficiently and Inaccurately Labeled Samples,"IEEE Transactions on Instrumentation andMeasurement,vol.70,pp.1-12,2021,Art no.2509912.. Parameters in the method of the present invention are consistent with parameters in the reference article. Table 1 shows the results of the ethanol concentration predictions for the eighth fermentation run using nine methods. It can be seen that the prediction accuracy of the proposed method is far better than the best-performing supervisory ladder network (high decision coefficient, small root mean square error of prediction) when only the marked samples are used. Due to differences in the operating environment, there is no guarantee that all unlabeled samples are generated from the same distribution as the labeled samples. All the virtual memory samples are generated from the same distribution as the component concentration, so the proposed method is superior to the semi-supervised ladder network method.
Table 1 comparison of predicted results based on eighth batch of samples
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (4)
1. A spectral data enhancement method based on conditional variation self-coding is characterized by comprising the following steps of: the method comprises the following steps:
Step one: preprocessing the collected historical data;
Measuring the component concentration of the product by using an instrument to obtain a labeled training set comprising a labeled spectrum X L＝[x1;x2;xi;…;xL and a corresponding component concentration Y L, drawing a trend chart of the obtained component concentration of the product, and deleting the spectrum and the component content which do not accord with the process trend from the training set; then, the clean training sets X L and Y L are subjected to standardization treatment so as to weaken the influence of external interference on the spectrum;
Step two: training a condition-variable automatic encoder to generate virtual samples;
Establishing a conditional variation self-encoder based on an encoder q (z|x, Y) and a decoder p (x|z, Y), setting a virtual component content Y U in a modeling set to generate a virtual spectrum X U, wherein X and Y are a spectrum and a component concentration of a training set respectively, p represents a probability distribution function, and z represents a latent variable;
generating a similar sample of representative marker samples, written as p (x|y)
p(x|y)＝∫p(z)p(x|y,z)dz
The probability model, that is, maximizing the collected data, is built as:
wherein logp (x|y) is further decomposed into:
Wherein L ELBO is the variation lower bound;
Since the value of the non-negativity of the KL divergence is greater than or equal to 0, there are
logp(x|y)≥LELBO
Wherein L ELBO is further decomposed into:
LELBO＝-KL(q(z|x,y)||p(z|y))+∫q(z|x,y)logp(x|z,y)dz
step three: training a semi-supervised ladder network to construct a calibration model;
Step four: the component content is detected on line by using a calibration model constructed by a semi-supervision ladder network;
and inputting the online spectrum sample into a semi-supervised ladder network spectrum calibration model guided by an established condition variation self-encoder, and detecting the component content online in real time.
2. A method of enhancing spectral data based on conditional variation self-encoding as claimed in claim 1, wherein: in the first step, the standardized processing formula of training sets X L and Y L is as follows:
wherein: x i,std represents the absorbance of the ith spectral data subjected to normalization treatment; x i represents the absorbance of the ith spectral data wave; m represents the average value of absorbance of the spectrum data with the label; s represents the standard deviation of absorbance of the spectral data with the label.
3. A method of enhancing spectral data based on conditional variation self-encoding as claimed in claim 1, wherein: the content Y U of the virtual component in the second step is within the range of Y L, the step length is set, and the number of generated virtual spectrums is more than 10 times of the data quantity with labels.
4. A method of enhancing spectral data based on conditional variation self-encoding as claimed in claim 1, wherein: and thirdly, inputting a virtual sample X U generated by using the condition variation self-encoder into an unlabeled part of the semi-supervised ladder network, inputting real marked samples X L and Y L into a labeled part of the semi-supervised ladder network, and performing semi-supervised learning by adopting the semi-supervised ladder network to obtain a spectrum calibration model of the semi-supervised ladder network guided by the condition variation self-encoder.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111546557.1A CN114239399B (en) | 2021-12-17 | 2021-12-17 | Spectral data enhancement method based on conditional variation self-coding |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111546557.1A CN114239399B (en) | 2021-12-17 | 2021-12-17 | Spectral data enhancement method based on conditional variation self-coding |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114239399A CN114239399A (en) | 2022-03-25 |
| CN114239399B true CN114239399B (en) | 2024-09-06 |
Family
ID=80757419
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111546557.1A Active CN114239399B (en) | 2021-12-17 | 2021-12-17 | Spectral data enhancement method based on conditional variation self-coding |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114239399B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110689086A (en) * | 2019-10-08 | 2020-01-14 | 郑州轻工业学院 | Semi-supervised high-resolution remote sensing image scene classification method based on generating countermeasure network |
| CN112990385A (en) * | 2021-05-17 | 2021-06-18 | 南京航空航天大学 | Active crowdsourcing image learning method based on semi-supervised variational self-encoder |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8831936B2 (en) * | 2008-05-29 | 2014-09-09 | Qualcomm Incorporated | Systems, methods, apparatus, and computer program products for speech signal processing using spectral contrast enhancement |
| US11003995B2 (en) * | 2017-05-19 | 2021-05-11 | Huawei Technologies Co., Ltd. | Semi-supervised regression with generative adversarial networks |
| CN113361481A (en) * | 2021-07-06 | 2021-09-07 | 北京明略昭辉科技有限公司 | Hyperspectral image classification method, device, equipment and storage medium |
-
2021
- 2021-12-17 CN CN202111546557.1A patent/CN114239399B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110689086A (en) * | 2019-10-08 | 2020-01-14 | 郑州轻工业学院 | Semi-supervised high-resolution remote sensing image scene classification method based on generating countermeasure network |
| CN112990385A (en) * | 2021-05-17 | 2021-06-18 | 南京航空航天大学 | Active crowdsourcing image learning method based on semi-supervised variational self-encoder |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114239399A (en) | 2022-03-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105158200B (en) | A kind of modeling method for improving the Qualitative Analysis of Near Infrared Spectroscopy degree of accuracy | |
| CN104020127A (en) | Method for rapidly measuring inorganic element in tobacco by near infrared spectrum | |
| Cetó et al. | Cava wine authentication employing a voltammetric electronic tongue | |
| CN102876816A (en) | Fermentation process statue monitoring and controlling method based on multi-sensor information fusion | |
| BR102017017235A2 (en) | ANALYSIS OF CHEMICAL AND PHYSICAL ATTRIBUTES OF SOIL FERTILITY BY VIS-NIR SPECTROSCOPY FOR USE IN LARGE SCALE ROUTINE | |
| Liu et al. | A novel hybrid model to predict dissolved oxygen for efficient water quality in intensive aquaculture | |
| CN117805045A (en) | Aquaculture online monitoring system and method based on image processing technology | |
| Mu et al. | Developing a conditional variational autoencoder to guide spectral data augmentation for calibration modeling | |
| CN116187861A (en) | Isotope-based water quality traceability monitoring method and related device | |
| CN112630180B (en) | Ultraviolet/visible light absorption spectrum model for detecting concentration of organophosphorus pesticide in water body | |
| CN109669017B (en) | Refinery distillation tower top cut water ion concentration prediction method based on deep learning | |
| Kaplun et al. | An intelligent agriculture management system for rainfall prediction and fruit health monitoring | |
| Quintero et al. | A novel data processing technique for expert resonant nano-pillars transducers: A case study measuring ethanol in water and wine liquid matrices | |
| CN114239399B (en) | Spectral data enhancement method based on conditional variation self-coding | |
| Marcon et al. | A statistical test for Ripley’s K function rejection of poisson null hypothesis | |
| CN112651173B (en) | A non-destructive testing method and generalizable system for agricultural product quality based on cross-domain spectral information | |
| CN106529680B (en) | A kind of multiple dimensioned extreme learning machine integrated modelling approach based on empirical mode decomposition | |
| Liu et al. | Incremental Support Vector Machine Combined with Ultraviolet‐Visible Spectroscopy for Rapid Discriminant Analysis of Red Wine | |
| CN114330114A (en) | Beryllium bronze alloy corrosion rate prediction method based on quantum support vector machine | |
| Jia et al. | Semi‐supervised kernel partial least squares fault detection and identification approach with application to HGPWLTP | |
| CN116884536B (en) | Automatic optimization method and system for production formula of industrial waste residue bricks | |
| US20230419174A1 (en) | Calibration device, calibration method, and computer-readable recording medium | |
| CN113970528B (en) | Textile component mixing method based on complete constraint least square method | |
| CN117589697A (en) | A method for detecting pH value of plant soil | |
| Liu et al. | Genetic algorithm‐based semisupervised convolutional neural network for real‐time monitoring of Escherichia coli fermentation of recombinant protein production using a Raman sensor |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |