CN108133313B - Artificial intelligent sensory evaluation food flavor system and construction method thereof - Google Patents
Artificial intelligent sensory evaluation food flavor system and construction method thereof Download PDFInfo
- Publication number
- CN108133313B CN108133313B CN201711358141.0A CN201711358141A CN108133313B CN 108133313 B CN108133313 B CN 108133313B CN 201711358141 A CN201711358141 A CN 201711358141A CN 108133313 B CN108133313 B CN 108133313B
- Authority
- CN
- China
- Prior art keywords
- data
- food
- sensory evaluation
- artificial
- flavor
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N30/00—Investigating or analysing materials by separation into components using adsorption, absorption or similar phenomena or using ion-exchange, e.g. chromatography or field flow fractionation
- G01N30/02—Column chromatography
Landscapes
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Development Economics (AREA)
- Physics & Mathematics (AREA)
- Entrepreneurship & Innovation (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- Educational Administration (AREA)
- Health & Medical Sciences (AREA)
- Marketing (AREA)
- Pathology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Analytical Chemistry (AREA)
- Immunology (AREA)
- General Health & Medical Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Biochemistry (AREA)
- Chemical & Material Sciences (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Other Investigation Or Analysis Of Materials By Electrical Means (AREA)
- Fats And Perfumes (AREA)
Abstract
The invention provides an artificial intelligence sensory evaluation food flavor system and a construction method thereof, wherein the system comprises a component sensory detection system, a computer intelligence analysis system and a food smell database, wherein the component sensory detection system consists of GC-MS and GC-O, the GC-MS is used for obtaining original smell characteristic data of food, and the GC-O is used for obtaining artificial sensory evaluation data of food smell; the computer intelligent analysis system is used for preprocessing the original smell characteristic data and the artificial sensory evaluation data, extracting the characteristics, and performing dimensionality reduction and clustering analysis on the original data; the food smell database is various data which are formed by performing fitting analysis on smell characteristic original data and artificial sensory evaluation data obtained by a sensory detection system through a computer and associating the volatile components of the food by combining the established smell original data and the artificial evaluation data. The invention integrates the instrument analysis and the human body sensory evaluation, overcomes the defects of one-sidedness and manual repeatability detection of the existing food flavor detection method, and has more comprehensive and reliable evaluation result.
Description
Technical Field
The invention relates to a system for evaluating food flavor, in particular to a system for evaluating food flavor based on artificial intelligence construction.
Background
As a source and a motive force of original innovation of the food industry, the flavor (smell) of the food is an important characteristic reflecting the quality of the food, and the quick and comprehensive evaluation and the material basis and the change rule thereof are worthy of deep research. In China, particularly Guangdong province, sensory flavor evaluation has a long history, and the method is particularly widely applied to production, processing and circulation management of traditional preference foods such as white spirit, tea, tobacco and the like. The flavor sensory evaluation technology mainly goes through three stages, namely: starting from the evaluation of a manager; evaluating professional sensory evaluation group products into a main body, crossing and applying multiple disciplines, and standardizing sensory evaluation activities; sensory analysis is combined with physical and chemical analysis, and the sensory evaluation is assisted by instrument measurement. After the twenty-first century, with the development of information technology, life science and instrument analysis technology, sensory science and technology are crossed with a plurality of subjects, and the development trend of man-machine integration is shown. Among them, sensor technologies for simulating human sense organs such as an electronic nose and an electronic tongue have been developed. For example, the Chinese patent discloses 'an olfactory analog device and method' (02111936.5) and 'a portable electronic pen and a manufacturing method thereof', which provide a sensor technology for simulating human sense organs.
However, neither artificial sensory evaluation, instrumental analysis, or electronic nose simulating human senses, etc. can well meet the development requirements of the food industry. Although the artificial sensory evaluation can comprehensively evaluate the food flavor, the aim of controlling the product quality is difficult to achieve only by using professionals to carry out the artificial sensory evaluation along with the large-scale and integrated development of food production, meanwhile, the artificial sensory evaluation is influenced by the psychology and physiology of appraisers, human errors are inevitable, and the sensory evaluation cannot know the material basis influencing the food flavor; although the evaluation technologies such as chemical analysis and instrument analysis can accurately and quantitatively analyze main typical flavor characteristic substances in food, the comprehensive flavor evaluation is difficult to be carried out like the evaluation of professionals; although the electronic nose can analyze the odor of specific components, the electronic nose is limited by the development of sensor technology, and is difficult to perform satisfactory analysis and evaluation on trace components and complex odor, and lacks self-learning and comprehensive analysis capabilities.
In recent years, the application of artificial intelligence technology is rapidly developed, especially Alpha Go proposed by google corporation, further pushing the application of artificial intelligence technology to the peak, and introducing artificial intelligence technology into various industries. The Alpha Go technology fully embodies the rapid development of artificial intelligence technologies with self-learning characteristics, such as a deep learning technology, an enhanced learning technology and the like. In the face of the trend, if an artificial intelligence self-learning technology can be introduced into the field of food flavor evaluation, multi-source and multi-mode data including artificial sensory evaluation data, instrument analysis data, image data and the like are better fused, the comprehensiveness of artificial sensory evaluation and the accuracy of instrument analysis are combined, the limitations of an electronic nose and an electronic tongue which mainly adopt a sensor technology are overcome, the objectivity and comprehensiveness of artificial sensory evaluation are improved, and a novel self-learning artificial intelligence sensory system for evaluating food flavor is constructed, so that the novel self-learning artificial intelligence sensory evaluation system not only has the inevitable trend of technical development and application requirements, but also has important scientific value and application prospects.
Disclosure of Invention
The invention aims to provide an artificial intelligence sensory evaluation food flavor system, which adopts an analytical instrument and human body sensory evaluation food flavor, the evaluation data is deeply learned in the fields of biological characteristic identification and food physical property analysis through a combined computer to form a food flavor autonomous evaluation model, and the artificial intelligence sensory evaluation food flavor (odor) system is constructed to evaluate the flavor of volatile food.
To achieve the above object, the embodiments of the present invention are: an artificial intelligent sensory evaluation food flavor system comprises a component sensory detection system, a computer intelligent analysis system and a food odor database,
the component sensory detection system consists of GC-MS and GC-O, wherein the GC-MS is used for obtaining original smell characteristic data (chromatogram and mass spectrum data of the content of each volatile component and the chemical structure of the food) of the food, and the GC-O is used for obtaining artificial sensory evaluation data (smell intensity and smell characteristic) of the food smell;
the computer intelligent analysis system is used for preprocessing the odor characteristic original data and the artificial sensory evaluation data obtained by the sensory detection system, extracting the characteristics, and performing dimensionality reduction and clustering analysis on the data;
the food odor database is formed by fitting and analyzing odor characteristic original data and artificial sensory evaluation data obtained by a sensory detection system through a computer artificial intelligent algorithm, eliminating error points, finding out association and rules of the data, and establishing various data of mutual association of food volatile components by combining the odor original data and the artificial evaluation data.
As a specific implementation mode, the computer intelligent analysis system adopts a multi-mode and multi-task deep learning framework M3And TDN, wherein the multi-mode refers to multi-mode of data, including chromatographic data, mass spectrum data and food appearance image data, and the multi-task refers to a plurality of targets to be learned simultaneously, including flavor type, flavor intensity and flavor amplitude.
In a specific embodiment, the food odor database comprises chromatographic information, mass spectrum information, aroma characteristics, aroma intensity and substance images of volatile components.
The invention also provides a construction method of the system, which comprises the following steps:
(1) the method comprises the following steps that a food sample enters a GC (gas chromatography), after separation through a capillary column, an outflow component is divided into two paths by a flow dividing valve, one path enters a chemical detector (FID or MS) to obtain original smell characteristic data (namely the content of each volatile component of the food and the chromatogram and mass spectrum data of a chemical structure), which are formed by original chromatogram and mass spectrum data, and the other path enters an sniffing port (O) through a special transmission pipeline to obtain artificial sensory evaluation data;
(2) fitting and analyzing the original odor characteristic data and the artificial sensory evaluation data (content, aroma intensity, aroma type and the like) through computer software, eliminating error points, and finding out the association and the rule of the data, thereby establishing a food odor database combining the original odor characteristic data and the artificial sensory evaluation data;
(3) normalizing and quantizing the original odor characteristic data and the artificial sensory evaluation data in the food odor database to obtain high-dimensional flavor data, and performing characteristic extraction on the high-dimensional flavor data by adopting a depth limited Boltzmann machine;
(4) training the high-dimensional multi-source feature data after feature extraction to construct a multi-modal and multi-task deep learning framework M3And TDN, on the basis of feature extraction, respectively adopting a multi-layer LSTM and a multi-layer restricted Boltzmann machine training and judging model, thereby establishing an artificial intelligence flavor evaluation system of the food.
In the step (1), the manual evaluation data is obtained by entering the sample into a sniffing port, recording manual knob electronic signals while sniffing by a person, and processing the recorded data result by a computer to form a spectrogram or convert information data.
In another embodiment, in step (3), the original odor feature data and the artificial sensory evaluation data obtained by machine judgment and artificial interpretation in the food database may be used as input data for feature processing, and the feature extraction and analysis may be performed by applying the LDA and PCA learning classical algorithms.
The verification (use) method of the invention is that the same type of samples in the food smell database are selected for pretreatment, and are analyzed and detected by a GC-MS instrument to obtain the original data of the chromatogram and the mass spectrum, and the original data of the chromatogram and the mass spectrum are introduced into the artificial intelligent flavor evaluation system, so that the system can automatically generate the flavor evaluation on the food.
The invention combines the instrument analysis and the human body sensory evaluation into a more scientific and effective evaluation mode, not only can reflect the visual perception of the human body to the food flavor, but also embodies the advantage that the analysis instrument obtains accurate detection data. The method not only overcomes the defects of one-sidedness and manual repeatability detection of the existing food flavor detection method, but also has more comprehensive and reliable evaluation results.
Drawings
FIG. 1 is a diagram of a system for sensory evaluation of flavor in foods using artificial intelligence according to the present invention;
FIG. 2 is a schematic flow chart of the study protocol of the present invention;
FIG. 3 is a graph showing the relationship between the peak area of the GC spectrum and the ethanol content in example 3 of the present invention;
FIG. 4 is a graph showing the relationship between the aroma score and the content of the ethanol solution in example 3 of the present invention.
Detailed Description
The present invention will be described in detail with reference to the accompanying drawings and examples.
Example 1: artificial intelligence sensory evaluation food flavor system and construction method thereof
As shown in figure 1, the artificial intelligence sensory evaluation food flavor system consists of an ingredient sensory detection system, a computer intelligence analysis system and a food smell database, wherein the ingredient sensory detection system consists of a GC-MS (gas chromatography-mass spectrometry detector) and a GC-O (gas chromatography-olfaction detector), the GC-MS is used for obtaining original smell characteristic data of a food sample, which consists of chromatographic and mass spectrometry data, and the GC-O is used for obtaining artificial sensory evaluation data of the smell of the food sample; the computer intelligent analysis system is used for preprocessing and extracting the odor characteristic original data and the artificial sensory evaluation data obtained by the sensory detection system, and performing dimensionality reduction and clustering analysis on the original data; the food odor database is characterized in that a computer artificial intelligence algorithm is used for carrying out fitting analysis on odor characteristic original data and artificial sensory evaluation data obtained by a sensory detection system, error points are eliminated, association and rules of the data are found, and established various data of food volatile components, which are combined by the odor original data and the artificial evaluation data, are associated with each other, comprise chromatographic information, mass spectrum information, aroma characteristics, aroma intensity, substance images of food samples and the like of the volatile components of the food samples.
The construction method of the artificial intelligence sensory evaluation food flavor system comprises the following steps:
1. and establishing a food smell database.
Adopting a proper pretreatment means and an extraction and separation method to carry out pretreatment and enrichment on odor components in the food raw materials; establishing a GC-MS-O system (gas chromatography-mass spectrometry-olfaction measurement system) for olfaction artificial sensory evaluation: the method comprises the steps of enabling a sample to enter a GC, separating the sample through a capillary column, dividing an outflow component into two paths by a flow dividing valve, enabling one path of the outflow component to enter a chemical detector (FID or MS), qualitatively judging each volatile component through a TIC (molecular ion peak) and a mass spectrum (fragment ion peak), and accurately quantifying the volatile component through a standard internal standard method to obtain an evaluation result of an analytical instrument on food smell. Converting the compound characteristic spectrum into original data of the chromatogram and the mass spectrum of each substance, namely original data of odor characteristics, and using the original data as odor identification data of a computer for a specific substance; and the other path enters a sniffing port through a special transmission pipeline, a professional smeller performs sniffing detection on the separated volatile components, and meanwhile, manual knob electronic signal recording is performed, and the aroma characteristics and the aroma intensity of each component are recorded, so that the manual evaluation data of human body senses on food aroma are obtained. Fitting and analyzing the content, aroma intensity and aroma type of the corresponding substances through computer software, eliminating error points, finding out the association and the rule of data, combining the original data of the aroma characteristics detected by an instrument and the artificial sensory evaluation data as comprehensive basic data of an artificial intelligent sensory evaluation food flavor system, and establishing a food aroma database.
2. Computer odor data preprocessing and feature extraction and evaluation model establishment
Multi-modal and multi-task deep learning framework M for combining food odor characteristic data detected by construction instrument with artificial sensory evaluation data3TDN to meet the requirement of intelligent flavor evaluation. In specific implementation, the collected odor characteristic raw data is used
Raw input data as structural components of a substance
. Wherein
Is the data of the color spectrum,
is mass spectral data, subscript
Corresponding to characteristic odor. Chromatographic data
Belongs to a time series signal, the wave crest of the time series signal corresponds to the corresponding substance component, and the area of the wave crest corresponds to the relative content of the corresponding substance component. Meanwhile, corresponding mass spectrum data is arranged for each peak
Is a heterogeneous mass/nuclear ratio（M/Z）The molecular structure of the corresponding substance component can be searched according to the existing spectrogram database. The human nose smells the sensory data (namely the artificial evaluation data) as the output data to be matched and judged
. Wherein the content of the first and second substances,
is a human sensory evaluationJudging the types of the obtained flavors, such as lemon flavor, woody flavor, flower flavor, gum flavor and the like,
is the judgment of strength made by human sensory evaluation, such as "micro" in slight peppery,
is the judgment of flavor amplitude, namely flavor duration, made by human sensory evaluation. The method specifically comprises the following steps: data preprocessing, feature extraction and deep learning model construction.
(1) And (4) preprocessing data. According to the total amount of sample substances and related instrument parameters in instrument analysis, normalizing and quantizing the acquired chromatogram and mass spectrum data (food odor characteristic data) to respectively obtain normalized vectors of substance contents
(wherein
Is a normalized value of the relative content of the kth material component measured in the chromatogram) and a normalized vector of the material component
(wherein
Is the ratio of the k-th substance component to the substance/nucleus: (M/Z) Molecules or ions ordered as p correspond to normalized values of intensity on the mass spectrum). After normalization, the flavor data obtained by one test can be expressed as a high-dimensional vector:
。
(2) and (5) feature extraction. The feature extraction is carried out by comparing and analyzing the two modes. First modeAnd carrying out traditional machine judgment and manual interpretation on the mass spectrum and the chromatographic data to obtain original data (including odor characteristic original data and artificial sensory evaluation data) of material components, material contents and the like, taking the original data as input data of characteristic processing, and then carrying out characteristic extraction analysis by applying machine learning classical algorithms such as LDA, PCA and the like. The second mode is to normalize the high-dimensional flavor data obtained by the signal data of chromatogram and mass spectrum after data preprocessing
And as input data, feature extraction and learning are carried out on the high-dimensional flavor data by adopting a depth-limited Boltzmann machine.
(3) And (5) constructing a deep learning judgment model. Training the odor high-dimensional multi-source characteristic data, and constructing a new multi-modal and multi-task deep learning framework M on the basis of the existing deep learning models such as CNN (convolutional neural network), LSTM (least squares), and the like3And (5) TDN. On the basis of feature extraction, a training evaluation model is respectively built by adopting a multilayer LSTM mechanism and a multilayer restricted Boltzmann mechanism. The LSTM is adopted in the depth structure, so that the correlation of different substance components in the formation of flavor characteristics is considered; the multi-mode refers to a multi-mode of data, including chromatographic data, mass spectrum data, food appearance image data and the like; multitasking refers to the simultaneous learning of multiple objectives of flavor type, flavor intensity, flavor breadth, etc.
Example 2: and (3) carrying out flavor evaluation on the Xinhui dried orange peels in different storage years by using an artificial intelligent sensory evaluation food flavor system.
The dried orange peel is used as a traditional Chinese medicine for regulating the flow of qi, the theory of 'good old people' is provided from ancient times, the volatile aroma is prominent, the content of volatile components and the aroma characteristics can be changed along with the extension of storage years, and an artificial intelligent evaluation system for the smell of the dried orange peel is built by establishing a database of the volatile components of the dried orange peel, so that the dried orange peel can be identified in different years and production places. In addition, the evaluation system can correlate the flavors of corresponding samples by comparing GC-MS data of the dried orange peel samples, so that labor force is saved, and the flavor data of the samples can be more accurately obtained.
(1) Data acquisition
As shown in fig. 1, the odor substances of the dried orange peel are researched, and the new-party dried orange peel of different years is selected as the experimental raw material, wherein the years of the dried orange peel comprise 1978, 1980, 1982, 1997, 1999, 2000, 2001, 2004, 2009, 2010 and 2012. Samples were prepared by ignoring their years and were numbered 1-11 for each sample. The samples were crushed with a grinder, 2 g of each sample was weighed, and a 75 μm CAR/PDMS needle was subjected to headspace-solid phase microextraction at 90 ℃ for 30min and resolved at the injection port for 10 min.
The GC-MS selects a DB-5 nonpolar elastic quartz capillary column, the temperature rise program is that the temperature of a sample inlet is 260 ℃, the flow rate is 1.0ml/min, and the carrier gas is high-purity helium (more than 99.999%); the initial temperature is 70 ℃, the temperature is kept for 2min, the temperature is raised to 210 ℃ at the speed of 4 ℃/min, the temperature is kept for 10min, and the split ratio is 50: 1. Mass spectrum conditions: the ion source was 230 ℃, the electron energy was 70eV, and the MS quadrupole temperature was 150 ℃.
The volatile components enter a mass spectrum end after being separated by a gas chromatographic column, and after multiple mass spectrum analysis, the volatile component types, the peak-appearing time and the relative content of the citrus reticulata blanco in each year, and original data including a chromatogram and a mass spectrum are obtained. The data are shown in Table 1.
(2) Data preprocessing and comparison
Arranging the data into computer recognizable form, numbering the pericarpium Citri Tangerinae and volatile components contained in all the pericarpium Citri Tangerinae in each year, such as the pericarpium Citri Tangerinae data No. 1978000 in 1978, and recording the original gas chromatogram data of the sample, including time and intensity of each point in the TIC chart, and peak area and peak height corresponding to each chromatographic peak. Raw mass spectral data for all sample peak species are aggregated, including the mass-to-charge ratio and intensity of each fragment ion of the species. Determining the qualitative and quantitative volatile components of the dried orange peel through the original data, and importing the data into a food smell artificial intelligence evaluation system to obtain the aroma information of the volatile components of the dried orange peel. The results are shown in Table 2.
(3) Flavor evaluation and proofreading
And performing GC-O detection on the volatile components of the dried orange peel by a fragrance smelling teacher, and performing sensory evaluation on the flavor of the dried orange peel. The method comprises the following steps:
taking 2 g of samples in each year, crushing by a grinder, adsorbing at 90 ℃ for 30min, extracting by a 75-micrometer CAR/PDMS extraction needle for 30min, and analyzing by a sample inlet for 10 min.
Chromatographic conditions are as follows: the chromatographic column is a DB-5 nonpolar elastic quartz capillary column, the programmed temperature rise is that the temperature of a sample inlet is 260 ℃, the flow rate is 1.0ml/min, and the carrier gas is high-purity helium (more than 99.999 percent); the initial temperature of the column is 70 ℃, the column is kept for 2min, the temperature is raised to 210 ℃ at the speed of 4 ℃/min, the column is kept for 10min, and the split ratio is 50: 1.
Mass spectrum conditions: the ion source was 230 ℃, the electron energy was 70eV, and the MS quadrupole temperature was 150 ℃.
A smell detector: the heating temperature of the ODP is 270 ℃, the tail blowing air flow is 50mL/min, and the humidifier humidifies the tail blowing air at the sniffing outlet so as to reduce the damage of dry air to the nasal mucosa. The early-stage training of the sniffer is used for carrying out experiments, and the sniffer records the time of smelling the odor, the fragrance characteristic and the fragrance intensity in the smell detection port. And table 4 shows the comparison of the evaluation results of the professional fragrance smelling teacher and the artificial intelligence evaluation system, and the accuracy of the professional fragrance smelling teacher is 100%.
Example 3: the flavor evaluation was performed on ethanol solutions of different concentrations.
(1) Establishing a sensory evaluation database and an intelligent flavor evaluation model:
at room temperature of 20 ℃, 21 clean and dry 100ml volumetric flasks are taken, 0, 5, 10, 15, 20, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, 85, 90, 95 and 100ml of absolute ethyl alcohol is accurately measured and placed in the 100ml volumetric flasks, and water is added to the volumetric flasks to fix the volume, so that 21 ethanol solution standard samples with the concentration of 0%, 5%, 10%, 15%, 20%, 25%, 30%, 35%, 40%, 45%, 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, 95% and 100% are prepared.
The 21 ethanol solution standard samples are used for training scent smellers of a sensory evaluation group by a scent smelling paper method to determine preliminary artificial sensory evaluation. The method is referred to GB/T22366-2008 as follows: providing the ethanol solution with one concentration together with two standard samples of the ethanol solution with the same other concentrations to a fragrance smelling teacher, and selecting a sample with a concentration different from the other two samples from the fragrance smelling teacher. The incense smelling teacher is trained in this way for 2 weeks, so that the incense smelling teacher is familiar with the discrimination ability.
And (3) feeding 21 ethanol solution standard samples into a GC, separating by a capillary column, dividing the outflow components into two paths by a flow dividing valve, feeding one path into a chemical detector MS and feeding the other path into a sniffing port through a special transmission pipeline, and sniffing by a smelling engineer and simultaneously recording a manual knob electronic signal. The GC-MS conditions were as follows:
chromatographic conditions are as follows: the chromatographic column is a DB-1 type polar elastic quartz capillary column. The temperature programming is that the temperature of a sample inlet is 250 ℃, the flow rate is 1.0ml/min, and the carrier gas is high-purity helium (more than 99.999 percent); the initial temperature of the column is 50 ℃, the column is kept for 1min, the temperature is raised to 110 ℃ at the speed of 15 ℃/min, the column is kept for 1min, and the split ratio is 8: 1.
Mass spectrum conditions: the interface temperature is 230 ℃, the ion source electron energy is 70eV, the electron multiplier high voltage is 1.0 kV, the mass number scanning range is 20-100, and the sampling rate is 0.2 s.
A smell detector: the heating temperature of the ODP is 150 ℃, the tail blowing air flow is 50mL/min, and the humidifier humidifies the tail blowing air at the sniffing outlet so as to reduce the damage of dry air to the nasal mucosa.
Recording data results (aroma characteristics and intensity), establishing a composition and relative content relation database (such as a table 5 and a figure 3) by combining aroma information data obtained by smelling of a smeller and data derived from GC analysis, and constructing an intelligent flavor evaluation model (such as a table 6 and a figure 4).
And (3) grading setting: the alcohol fragrance is divided into 6 odor grades of no (0-10), light fragrance (10-20), faint scent (20-30), thick fragrance (30-40), pungent smell (40-50) and pungent smell (50-60), and the fragrance intensity score interval is 0-60.
(2) Verification of sensory evaluation model
5 ethanol samples to be tested were prepared at random concentrations: in the field of the fragrance smeller, a clean and dry 100ml measuring cylinder is taken, absolute ethyl alcohol with any amount is poured at will, the volume of the absolute ethyl alcohol is recorded, the measuring cylinder is moved into a clean and dry 100ml measuring cylinder, the volume is determined by water, the content of the ethyl alcohol is recorded but not marked, the operation is repeated for 5 times, and the numbers of all solutions are given by numbers 1-5.
Evaluation by fragrance smellers: the 5 samples were repeated three times each, all samples were sequentially scrambled, re-marked with letters a-o, submitted to independent sniffing by a fragrance smeller, describing fragrance type and intensity, and the results were recorded.
The 5 ethanol samples to be tested are subjected to GC-MS analysis and GC-O analysis by the following method:
chromatographic conditions are as follows: the chromatographic column is a DB-1 type polar elastic quartz capillary column. The temperature programming is that the temperature of a sample inlet is 250 ℃, the flow rate is 1.0ml/min, and the carrier gas is high-purity helium (more than 99.999 percent); the initial temperature of the column is 50 ℃, the column is kept for 1min, the temperature is raised to 110 ℃ at the speed of 15 ℃/min, the column is kept for 1min, and the split ratio is 8: 1.
Mass spectrum conditions: the interface temperature is 230 ℃, the ion source electron energy is 70eV, the electron multiplier high voltage is 1.0 kV, the mass number scanning range is 20-100, and the sampling rate is 0.2 s.
A smell detector: the heating temperature of the ODP is 150 ℃, the tail blowing air flow is 50mL/min, and the humidifier humidifies the tail blowing air at the sniffing outlet so as to reduce the damage of dry air to the nasal mucosa.
And transferring the original data derived from the GC-MS into an established intelligent flavor evaluation model for calculation to obtain the data of the ethanol content and the aroma intensity. The results are shown in Table 7.
Claims (4)
1. An artificial intelligence sensory evaluation food flavor system is characterized by comprising a component sensory detection system, a computer intelligence analysis system and a food smell database,
the component sensory detection system consists of GC-MS and GC-O, wherein the GC-MS is used for obtaining original smell characteristic data of food, and the GC-O is used for obtaining artificial sensory evaluation data of food smell, namely smell intensity and smell characteristics;
the computer intelligent analysis system is used for preprocessing and extracting the odor characteristic original data and the artificial sensory evaluation data obtained by the sensory detection system, and performing dimensionality reduction and clustering analysis on the original data; the computer intelligent analysis system adopts a multi-mode and multi-task deep learning framework M3TDN, wherein the multi-mode refers to multi-mode of data, including chromatographic data, mass spectrum data and food appearance image data, and the multi-task refers to a plurality of targets to be learned simultaneously, including flavor type, flavor intensity and flavor amplitude;
data preprocessing: normalizing and quantizing the acquired chromatographic data and mass spectrum data according to the total amount of sample substances and related instrument parameters in instrument analysis, normalizing the normalized vector of the substance content and the normalized vector of the substance component respectively, and performing one-time test to obtain flavor data expressed as high-dimensional vectors;
feature extraction: the characteristic extraction is to compare and analyze two modes, wherein in the first mode, original data of material components and material contents, including odor characteristic original data and artificial sensory evaluation data, are obtained by carrying out traditional machine judgment and artificial interpretation on the mass spectrum data and the chromatographic data and are used as input data of characteristic processing, and then characteristic extraction analysis is carried out by applying LDA and PCA machine learning classical algorithm; in the second mode, high-dimensional flavor data, namely high-dimensional vectors, which are obtained by data preprocessing and normalization of mass spectrum data and chromatographic data are used as input data, and a depth-limited Boltzmann machine is adopted to perform feature extraction and learning on the high-dimensional flavor data;
the food odor database is formed by fitting and analyzing odor characteristic original data and artificial sensory evaluation data obtained by a sensory detection system through a computer artificial intelligent algorithm, eliminating error points, finding out association and rules of the data, and establishing various data of mutual association of food volatile components by combining the odor original data and the artificial evaluation data.
2. The artificial intelligence sensory evaluation food flavor system of claim 1, wherein the food smell database comprises chromatographic information, mass spectrum information, aroma characteristics, aroma intensity and substance images of volatile components.
3. A construction method of an artificial intelligence sensory evaluation food flavor system is characterized by comprising the following steps:
(1) the food sample enters a GC, after separation through a capillary column, the outflow component is divided into two paths by a flow dividing valve, one path enters a chemical detector FID or MS to obtain odor characteristic original data consisting of chromatogram and mass spectrum original data, and the other path enters an sniffing port O through a special transmission pipeline to obtain artificial sensory evaluation data;
(2) fitting and analyzing the original odor characteristic data and the artificial sensory evaluation data through computer software, eliminating error points, and finding out the association and the rule of the data, thereby establishing a food odor database combining the original odor characteristic data and the artificial sensory evaluation data; the original data of the odor characteristics and the artificial sensory evaluation data comprise content, aroma intensity and aroma type;
(3) normalizing and quantizing the original odor characteristic data and the artificial sensory evaluation data in the food odor database to obtain high-dimensional flavor data, and performing characteristic extraction on the high-dimensional flavor data by adopting a depth limited Boltzmann machine;
the first mode of the characteristic extraction method is that odor characteristic original data and artificial sensory evaluation data obtained by using machine judgment and artificial interpretation modes in a food database are used as input data of characteristic processing, and characteristic extraction analysis is carried out by applying LDA and PCA learning classical algorithm; in the second mode, a high-dimensional vector obtained by normalizing mass spectrum data and chromatographic data after data preprocessing is used as input data, and a depth-limited Boltzmann machine is adopted to perform feature extraction and learning on the high-dimensional flavor data;
(4) training the high-dimensional multi-source feature data after feature extraction to construct a multi-modal and multi-task deep learning framework M3And TDN, on the basis of feature extraction, respectively adopting a multi-layer LSTM and a multi-layer restricted Boltzmann machine training and judging model, thereby establishing an artificial intelligence flavor evaluation system of the food.
4. The method for constructing an artificial intelligence sensory evaluation food flavor system according to claim 3, wherein in the step (1), the artificial sensory evaluation data is obtained by entering a sample into a sniffing port, performing manual knob electronic signal recording while sniffing by a human nose, and processing the recorded data result by a computer to form a spectrogram or convert information data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| PCT/CN2018/000423 WO2019114052A1 (en) | 2017-11-12 | 2018-12-17 | Artificial intelligence-based food flavor sensory evaluation system and establishment method therefor |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711110672 | 2017-11-12 | ||
| CN2017111106728 | 2017-11-12 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108133313A CN108133313A (en) | 2018-06-08 |
| CN108133313B true CN108133313B (en) | 2021-07-20 |
Family
ID=62390497
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711358141.0A Active CN108133313B (en) | 2017-11-12 | 2017-12-17 | Artificial intelligent sensory evaluation food flavor system and construction method thereof |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN108133313B (en) |
| WO (1) | WO2019114052A1 (en) |
Families Citing this family (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108133313B (en) * | 2017-11-12 | 2021-07-20 | 华南农业大学 | Artificial intelligent sensory evaluation food flavor system and construction method thereof |
| CN109409635B (en) * | 2018-08-27 | 2021-08-20 | 中化现代农业有限公司 | Agricultural product sensory data processing and acquiring method, device and system |
| CN109374802B (en) * | 2018-09-06 | 2020-07-07 | 北京工商大学 | Method for distinguishing cut cheese grade based on identity characteristic flavor component |
| CN109709221A (en) * | 2018-12-14 | 2019-05-03 | 江苏恒顺醋业股份有限公司 | The lookup analysis method of smell substance in a kind of fermentation vinegar liquid based on GC-MS |
| JP6788204B2 (en) * | 2019-02-12 | 2020-11-25 | 曽田香料株式会社 | Evaluation method of volatile components and equipment used for it |
| CN110082458B (en) * | 2019-05-31 | 2022-05-03 | 北京工商大学 | Method for identifying correlation between cheddar cheese volatile substances and sensory quality grade |
| CN110596330A (en) * | 2019-10-16 | 2019-12-20 | 粤海永顺泰(广州)麦芽有限公司 | Special malt quality evaluation method |
| CN111260199A (en) * | 2020-01-10 | 2020-06-09 | 江西理工大学 | Artificial intelligent sensory evaluation food flavor system and construction method thereof |
| CN111474287A (en) * | 2020-03-19 | 2020-07-31 | 杭州憶盛医疗科技有限公司 | Computer-aided system and method for analyzing composition components of medicine |
| CN111693661A (en) * | 2020-06-11 | 2020-09-22 | 上海交通大学 | Intelligent sense organ based dry-cured ham flavor evaluation method |
| CN111695825A (en) * | 2020-06-16 | 2020-09-22 | 内蒙古蒙牛乳业（集团）股份有限公司 | Food flavor detection method and device and electronic equipment |
| US11881093B2 (en) | 2020-08-20 | 2024-01-23 | Denso International America, Inc. | Systems and methods for identifying smoking in vehicles |
| US11636870B2 (en) | 2020-08-20 | 2023-04-25 | Denso International America, Inc. | Smoking cessation systems and methods |
| US11760169B2 (en) | 2020-08-20 | 2023-09-19 | Denso International America, Inc. | Particulate control systems and methods for olfaction sensors |
| US11813926B2 (en) | 2020-08-20 | 2023-11-14 | Denso International America, Inc. | Binding agent and olfaction sensor |
| US11760170B2 (en) | 2020-08-20 | 2023-09-19 | Denso International America, Inc. | Olfaction sensor preservation systems and methods |
| US11828210B2 (en) | 2020-08-20 | 2023-11-28 | Denso International America, Inc. | Diagnostic systems and methods of vehicles using olfaction |
| US11932080B2 (en) | 2020-08-20 | 2024-03-19 | Denso International America, Inc. | Diagnostic and recirculation control systems and methods |
| JP7495059B2 (en) * | 2020-11-02 | 2024-06-04 | 株式会社島津製作所 | Gas recovery device for odor evaluation, odor evaluation method, and odor evaluation sample preparation device |
| CN112858650B (en) * | 2021-01-13 | 2023-08-18 | 上海应用技术大学 | Analysis and research method for improving automobile leather smell based on sigma-tau strength method |
| CN113504316B (en) * | 2021-05-31 | 2023-08-29 | 江苏中烟工业有限责任公司 | Method for evaluating main fragrance of Maillard reaction product by utilizing functional groups |
| CN114324716B (en) * | 2021-08-30 | 2023-06-30 | 华南农业大学 | Method for measuring oxidation rancidity degree of infant formula rice flour and application thereof |
| CN114487234A (en) * | 2022-02-11 | 2022-05-13 | 云南中烟工业有限责任公司 | Method for simultaneously predicting moisture content and smoke characteristic index judgment value in tobacco drying process based on HS-GC-MS |
| CN115047091A (en) * | 2022-04-06 | 2022-09-13 | 中华全国供销合作总社济南果品研究所 | Analysis method for aroma components in dried red date slices |
| WO2024038419A1 (en) * | 2022-08-19 | 2024-02-22 | The Live Green Group, Inc. | Plant-only cheese flavoring replacement system in food products |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102323385A (en) * | 2011-06-03 | 2012-01-18 | 上海应用技术学院 | Method for measuring smell threshold of volatile flavor compound and application thereof |
| CN106874948A (en) * | 2017-02-08 | 2017-06-20 | 武汉海卓科科技有限公司 | A kind of black smelly water automatic identification and appraisal procedure |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103343049B (en) * | 2009-01-13 | 2014-05-07 | 花王株式会社 | Fragrance composition |
| CN102135526B (en) * | 2010-12-10 | 2013-05-29 | 中国农业科学院农产品加工研究所 | Method for extracting characteristic flavor substance of roasted peanuts |
| CN103105440A (en) * | 2012-11-06 | 2013-05-15 | 天津科技大学 | Method for identifying volatile flavor substance in bolete |
| CN103487558B (en) * | 2013-07-30 | 2016-10-12 | 中国标准化研究院 | A kind of method detecting exceptional sample during the pattern recognition analysis applying intelligent sensory signal to carry out tea leaf quality |
| CN104316660B (en) * | 2014-10-27 | 2016-07-06 | 四川省农业科学院农产品加工研究所 | The method that quickly detection thick broad-bean sauce local flavor, grade and product are traced to the source |
| CN104316635A (en) * | 2014-10-28 | 2015-01-28 | 石河子大学 | Method for rapidly identifying flavor and quality of fruits |
| CN108133313B (en) * | 2017-11-12 | 2021-07-20 | 华南农业大学 | Artificial intelligent sensory evaluation food flavor system and construction method thereof |
-
2017
- 2017-12-17 CN CN201711358141.0A patent/CN108133313B/en active Active
-
2018
- 2018-12-17 WO PCT/CN2018/000423 patent/WO2019114052A1/en active Application Filing
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102323385A (en) * | 2011-06-03 | 2012-01-18 | 上海应用技术学院 | Method for measuring smell threshold of volatile flavor compound and application thereof |
| CN106874948A (en) * | 2017-02-08 | 2017-06-20 | 武汉海卓科科技有限公司 | A kind of black smelly water automatic identification and appraisal procedure |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108133313A (en) | 2018-06-08 |
| WO2019114052A1 (en) | 2019-06-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108133313B (en) | Artificial intelligent sensory evaluation food flavor system and construction method thereof | |
| CN106501470B (en) | Utilize the method for gustatory system and electronic nose association evaluation mustard thick chilli sauce flavor grade | |
| CN105138861B (en) | A kind of construction method of the rhubarb medicinal material true and false and base source kind class prediction model | |
| CN109254107B (en) | Rapid classification and identification method for citrus Pu' er tea | |
| CN106841083A (en) | Sesame oil quality detecting method based on near-infrared spectrum technique | |
| CN102778442B (en) | Method for rapidly identifying type of balsam material liquid for cigarette | |
| CN103134850A (en) | Tea quality rapid detection apparatus and detection method based on characteristic fragrance | |
| CN110307871B (en) | Method for rapidly detecting quality of traditional Chinese medicine decoction pieces by adopting electronic sensory fusion | |
| CN106841494A (en) | Plant otherness metabolin rapid screening method based on UPLC QTOF | |
| CN101261280A (en) | Traditional Chinese herb odor discriminate method based on bionic olfaction and its device | |
| CN102023137A (en) | Method for identifying white spirits | |
| CN106885851A (en) | A kind of black tea place of production method of discrimination based on chiral quantitative analysis tech | |
| Huang et al. | Scent profiling of Cymbidium ensifolium by electronic nose | |
| CN103115892A (en) | Method for preparing ginkgo leaf extract by using near infrared spectroscopy analysis technology | |
| CN109030611B (en) | Rapid and lossless green tea aroma fingerprint analysis and aroma type discrimination method based on ion mobility spectrometry | |
| CN111239316A (en) | Cigarette sensory quality analysis method based on smoke all-chemical component chromatographic data | |
| CN103308637B (en) | Gas chromatography-mass spectrometry method for identifying dalbergia odorifera and dalbergia tonkinensi | |
| CN112116964A (en) | Detection method for rapidly judging fruit producing area | |
| CN109784390B (en) | Artificial intelligent smell dynamic response atlas gas detection and identification method | |
| CN106770607B (en) | A method of identifying genuine-fake cigarette using HS-IMR-MS | |
| CN110687215A (en) | Method for predicting tea year by using metabonomics technology | |
| CN103293264A (en) | Method for distinguishing quality of yeast | |
| CN107462624A (en) | The rapid assay methods of main ester type compound content in a kind of white wine base liquor | |
| CN202794093U (en) | Device for fast detecting quality of baked food based on bionic olfaction | |
| CN111912823A (en) | Multi-component pesticide residue fluorescence detection analysis method |
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 | ||
| CB03 | Change of inventor or designer information |
Inventor after: Cao Yong Inventor after: Wen Linfeng Inventor after: Fu Jiangyan Inventor after: Liu Zhan Inventor after: Hu Yuanshu Inventor after: Qiu zhirou Inventor after: Chang Huiyou Inventor after: Chen Haoran Inventor after: Sun Shengwei Inventor after: Hu Yongjun Inventor after: Chen Tong Inventor after: Liu Guo Inventor after: Kan Qixin Inventor before: Cao Yong Inventor before: Qiu zhirou Inventor before: Chang Huiyou Inventor before: Chen Haoran Inventor before: Sun Shengwei Inventor before: Hu Yongjun Inventor before: Chen Tong |
|
| CB03 | Change of inventor or designer information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |