WO2021253874A1 - 卷烟主流烟气光谱数据的感官评价方法 - Google Patents

卷烟主流烟气光谱数据的感官评价方法 Download PDF

Info

Publication number
WO2021253874A1
WO2021253874A1 PCT/CN2021/079141 CN2021079141W WO2021253874A1 WO 2021253874 A1 WO2021253874 A1 WO 2021253874A1 CN 2021079141 W CN2021079141 W CN 2021079141W WO 2021253874 A1 WO2021253874 A1 WO 2021253874A1
Authority
WO
WIPO (PCT)
Prior art keywords
cigarette
mainstream smoke
sensory
deep
shallow
Prior art date
Application number
PCT/CN2021/079141
Other languages
English (en)
French (fr)
Inventor
范多青
王慧
李超
吴亿勤
秦云华
苏杨
刘巍
王璐
高文军
Original Assignee
云南中烟工业有限责任公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 云南中烟工业有限责任公司 filed Critical 云南中烟工业有限责任公司
Priority to EP21806940.9A priority Critical patent/EP3961189B1/en
Priority to JP2021568873A priority patent/JP7203252B2/ja
Priority to US17/595,852 priority patent/US11525774B2/en
Publication of WO2021253874A1 publication Critical patent/WO2021253874A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
    • G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
    • G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
    • G01N21/35—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light
    • G01N21/3504—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light for analysing gases, e.g. multi-gas analysis
    • 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
    • G06Q10/06395—Quality analysis or management
    • A—HUMAN NECESSITIES
    • A24—TOBACCO; CIGARS; CIGARETTES; SIMULATED SMOKING DEVICES; SMOKERS' REQUISITES
    • A24C—MACHINES FOR MAKING CIGARS OR CIGARETTES
    • A24C5/00—Making cigarettes; Making tipping materials for, or attaching filters or mouthpieces to, cigars or cigarettes
    • A24C5/32—Separating, ordering, counting or examining cigarettes; Regulating the feeding of tobacco according to rod or cigarette condition
    • A24C5/34—Examining cigarettes or the rod, e.g. for regulating the feeding of tobacco; Removing defective cigarettes
    • A24C5/3406—Controlling cigarette combustion
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
    • G01N33/0004—Gaseous mixtures, e.g. polluted air
    • G01N33/0009—General constructional details of gas analysers, e.g. portable test equipment
    • G01N33/0027—General constructional details of gas analysers, e.g. portable test equipment concerning the detector
    • 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
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/2433—Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/01—Arrangements or apparatus for facilitating the optical investigation
    • G01N21/03—Cuvette constructions
    • G01N21/05—Flow-through cuvettes
    • G01N2021/052—Tubular type; cavity type; multireflective
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/01—Arrangements or apparatus for facilitating the optical investigation
    • G01N21/03—Cuvette constructions
    • G01N21/031—Multipass arrangements
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2201/00—Features of devices classified in G01N21/00
    • G01N2201/12—Circuits of general importance; Signal processing
    • G01N2201/129—Using chemometrical methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching

Definitions

  • the invention relates to the technical field of tobacco product quality evaluation, in particular to a sensory evaluation method for cigarette mainstream smoke spectral data.
  • Cigarette is a special smoking consumer product.
  • the content of key components in mainstream smoke directly affects the quality and taste of cigarettes. How to effectively quantify the correlation between smoke components and sensory evaluation to achieve accurate product evaluation has become the tobacco industry Key technical bottlenecks that need to be resolved urgently.
  • the sensory evaluation method for mainstream cigarette smoke spectrum data is obtained through expert evaluation, that is, sensory evaluation is obtained by professional smoking cigarettes and then scoring the corresponding scoring items.
  • This sensory evaluation method relies on manual labor, and the manual evaluation is affected by the sensory evaluation personnel’s own factors and the external environment. This brings uncertain factors to the cigarette evaluation results.
  • the expert evaluation method is time-consuming, laborious and process-intensive. Problems such as cumbersomeness, low efficiency, and unstable sensory evaluation results. Therefore, there is an urgent need for a sensory evaluation method for mainstream cigarette smoke spectrum data.
  • the purpose of the present invention is to provide a sensory evaluation method for mainstream cigarette smoke spectrum data, to solve the above-mentioned problems in the prior art, and to improve the efficiency, accuracy and stability of sensory evaluation of mainstream cigarette smoke.
  • the present invention provides a sensory evaluation method for mainstream cigarette smoke spectrum data, which includes:
  • the enhancement processing on the cigarette mainstream smoke spectrum data of several cigarettes specifically includes:
  • the component disturbance processing is performed on the cigarette mainstream smoke spectrum data of each cigarette after the physical disturbance processing.
  • At least one method of second-order differentiation, Karl Norris derivative filter, multivariate scattering correction, and wavelet transformation is used to perform noise reduction processing on each of the cigarette mainstream smoke spectrum data from which abnormal data is eliminated.
  • the construction methods include:
  • each of the cigarette mainstream smoke spectrum data extracted by the feature peak is trained to obtain the first sensory classification model.
  • the sensory evaluation method for cigarette mainstream smoke spectral data as described above, wherein, preferably, the deep spatial features are extracted from the cigarette mainstream smoke spectral data of each cigarette from which the shallow spectral features have been extracted Specifically including:
  • deep spatial features are extracted from the cigarette mainstream smoke spectral data of each cigarette from which the shallow spectral features have been extracted.
  • Input a number of the deep spatial features extracted based on the deep residual convolutional neural network into a support vector machine in a stack manner to obtain a second sensory classification model
  • the best classification error is used as the first objective function to obtain the first best network parameters
  • the balance point of the convolution kernel size of the first optimal convolution kernel corresponding to the first optimal network parameter and the second optimal convolution kernel corresponding to the second optimal network parameter is selected as the depth residual
  • the final network parameters of the difference convolutional neural network is selected as the depth residual
  • Comprehensive sensory quality results including:
  • a weighted summation is performed on the shallow sensory quality results and the deep sensory quality results to obtain a comprehensive sensory quality result.
  • the present invention provides a sensory evaluation method for mainstream cigarette smoke spectral data.
  • the spectral features implicit in the mainstream cigarette smoke can be maximized under limited sample conditions And spatial feature paradigm, thereby effectively reducing the demand for training sample size; extracting shallow spectral features can provide guidance information for key components of complex systems for subsequent deep learning, and help improve the extraction accuracy of deep spatial features; Extracting deep spatial features can quickly learn effective deep feature representation from training data, and thereby enhance the feature information expression of abnormal samples and normal samples; the present invention extracts spectral features and spatial features separately from shallow to deep, and Through the fusion of the spectral-spatial classification framework, the sensory evaluation results of mainstream cigarette smoke are automatically and directly obtained, and accurate screening of unknowns in mainstream smoke is realized.
  • Fig. 1 is a flowchart of an embodiment of a sensory evaluation method for mainstream cigarette smoke spectrum data provided by the present invention.
  • a specific component when it is described that a specific component is located between the first component and the second component, there may or may not be an intermediate component between the specific component and the first component or the second component.
  • the specific component When it is described that a specific component is connected to another component, the specific component may be directly connected to the other component without an intervening component, or may not be directly connected to the other component but with an intervening component.
  • the sensory evaluation method of cigarette mainstream smoke spectrum data provided by this embodiment in the actual implementation process specifically includes:
  • Step S1 Perform enhancement processing on the mainstream cigarette smoke spectrum data of a number of cigarettes.
  • the mainstream cigarette smoke spectrum data includes mid-infrared spectrum data, which is specifically obtained based on the hollow-core waveguide two-dimensional infrared spectroscopy detection technology.
  • the hollow-core waveguide infrared spectroscopy technology as an infrared enhancement technology, is similar to conventional infrared spectroscopy technology.
  • the incident light source forms multiple reflections in the hollow core fiber to extend the optical path of the interaction between light and matter, which can more efficiently improve the system under test. Infrared absorption intensity, thereby reducing the detection limit, and improving the precision and accuracy of the analysis.
  • Enhancing the spectral data of mainstream cigarette smoke can reduce the need for the number of training samples from the perspective of global optimization, and can enhance the spatial paradigm of the spectrum of components related to mainstream smoke (for example, hollow-core waveguide two-dimensional infrared spectroscopy). Effectively reduce the risk of overfitting between the first sensory classification model and the second sensory classification model (to be described later). In this way, under limited sample conditions, the spectral features and spatial feature paradigm implicit in mainstream cigarette smoke can be maximized, and the second sensory classification model can deepen the existing data, thereby effectively reducing the amount of training samples. Demand.
  • the spatial paradigm of the spectrum of components related to mainstream smoke for example, hollow-core waveguide two-dimensional infrared spectroscopy.
  • the step S1 may specifically include:
  • Step S11 Perform horizontal inversion processing on the mainstream cigarette smoke spectrum data of each cigarette.
  • Step S12 Perform random cutting processing on the cigarette mainstream smoke spectrum data of each cigarette that has been horizontally flipped.
  • the present invention adopts the traditional image enhancement mode and performs operations such as horizontal flipping and random cropping on it, so as to improve the effectiveness and robustness of the second sensory classification model for spatial image recognition.
  • Step S13 Perform physical disturbance processing on the cigarette mainstream smoke spectrum data of each cigarette that has been randomly cut.
  • the two-dimensional spectrum information is not only affected by the material composition information, but also by its physical state. Therefore, by changing the physical state of the sample, more information about the hollow-core waveguide two-dimensional infrared spectrum paradigm will be effectively obtained, thereby improving the accuracy and effectiveness of deep learning.
  • the physical disturbance factors in mainstream cigarette smoke mainly include the effects of different physical conditions such as temperature, wind speed, pressure, and extraction method on the hollow-core waveguide two-dimensional infrared spectrum information of mainstream cigarette smoke.
  • Step S14 Perform component disturbance processing on the cigarette mainstream smoke spectrum data of each cigarette that has undergone physical disturbance processing.
  • the information changes of the two-dimensional infrared spectroscopy of the mainstream cigarette smoke in different component disturbance states can be determined.
  • the paradigm information of the actual sample hollow core waveguide two-dimensional infrared spectrum signal can be improved, and the subsequent classification and recognition performance of the subsequent hollow core waveguide two-dimensional infrared spectrum can be improved.
  • Step S2 extracting shallow spectral features from the cigarette mainstream smoke spectral data of each cigarette after data enhancement processing.
  • the step S2 may specifically include:
  • Step S21 Use the Hotelling T 2 statistic of the spectral vector to eliminate outlier data points in the mainstream cigarette smoke spectrum data of a number of cigarettes, so as to eliminate abnormal data in the mainstream cigarette smoke spectrum data.
  • Step S22 using at least one method of second-order differentiation, Karl Norris derivative filter, multiplicative scatter correction (MSC), and wavelet transform to denoise each of the cigarette mainstream smoke spectral data from which abnormal data has been eliminated deal with.
  • MSC multiplicative scatter correction
  • Noise reduction processing can reduce noise interference, make the characteristic peaks in the mainstream cigarette smoke spectrum data more obvious, and facilitate the extraction of characteristic peaks from the background in the mainstream cigarette smoke spectrum data, so as to improve the signal-to-noise ratio. Moreover, through data screening and noise reduction processing, it is convenient to guide subsequent spectral analysis methods to accurately extract the data characteristics of the substance to be tested.
  • Step S3 based on the cigarette mainstream smoke spectrum data of each cigarette from which the shallow spectrum characteristics have been extracted and the shallow spectrum characteristics, obtain the shallow sensory quality results of each of the cigarette mainstream smoke spectrum data.
  • the purpose of extracting the shallow spectral features is to extract the spectral features of the key components of the sample from the complex and fluctuating spectrum (for example, the two-dimensional infrared spectrum of a hollow core waveguide) signal, and to reduce the dimensionality of the spectrum.
  • each of the cigarette mainstream smoke spectral data from which the shallow spectral characteristics has been extracted is input into a pre-built first sensory classification model to obtain the shallow sensory quality results of each of the cigarette mainstream smoke spectral data.
  • the first sensory classification model is constructed based on Principal Component Analysis (PCA) combined with a non-linear support vector machine (SVM), and the method for constructing the first sensory classification model is specifically include:
  • each of the cigarette mainstream smoke spectrum data extracted by the feature peak is trained to obtain the first sensory classification model.
  • the classification result of the first sensory classification model includes at least good, medium and poor.
  • the classification results of the first sensory classification model are good, neutral, and poor. It should be noted that the present invention does not specifically limit the classification results and number of the first sensory classification model. By defining parameters and changing weights, Other classification results can be obtained.
  • the expert sensory evaluation score is used to supervise the first classification judgment value output by the first sensory classification model, so that the verification of the first sensory classification model can be realized. And update.
  • the first sensory classification model is trained by the following training method:
  • Step S4 extracting deep spatial features from the cigarette mainstream smoke spectral data of each cigarette from which the shallow spectral features have been extracted.
  • deep spatial features are extracted from the cigarette mainstream smoke spectral data of each cigarette from which the shallow spectral features have been extracted.
  • the spatial topology information of the original key components will cause distortion to a certain extent, and the deep residual convolutional neural network can deal with this change and learn effectively from the training data.
  • Step S5 Based on the cigarette mainstream smoke spectral data of each cigarette from which the deep spatial features have been extracted and the deep spatial features, the deep sensory quality results of the cigarette mainstream smoke spectral data are obtained.
  • the purpose of extracting deep spatial features is to use the deep residual convolutional neural network method to complete the extraction and enhancement of deep spatial features in a translation-invariant way.
  • the method for determining the network parameters of the deep residual convolutional neural network specifically includes:
  • the best classification error is used as the first objective function to obtain the first best network parameters
  • the balance point of the convolution kernel size of the first optimal convolution kernel corresponding to the first optimal network parameter and the second optimal convolution kernel corresponding to the second optimal network parameter is selected as the depth residual
  • the final network parameters of the difference convolutional neural network is selected as the depth residual
  • the step S5 may specifically include:
  • Step S51 Input a number of the deep spatial features extracted based on the deep residual convolutional neural network into a support vector machine in a stack to obtain a second sensory classification model.
  • the classification result of the second sensory classification model is the sensory score.
  • the classification result of the second sensory classification model is the sensory score. It should be noted that the present invention does not specifically limit the classification results and number of the second sensory classification model. Others can be obtained by defining parameters and changing weights. Classification results.
  • Step S52 inputting each of the cigarette mainstream smoke spectral data from which the deep spatial features have been extracted into the second sensory classification model to obtain a deep sensory quality result of each of the cigarette mainstream smoke spectral data.
  • the present invention introduces a deep learning strategy to simulate the sensory evaluation process of different experts.
  • the objective problem of analyzing the sensory data of a single cigarette through mainstream smoke is divided into a multi-classification problem configured with predictive models with different weights and parameters to simulate different experts.
  • the score information of the expert labels obtained by the evaluation of artificial experts is used to improve the supervision information of the deep learning model.
  • marking improvement the symbiotic relationship between the score information obtained through the second sensory classification model and the score information of the expert tags obtained through the evaluation of human experts is used to improve the deep learning training results that have been obtained.
  • the present invention also selects standard cigarette samples with different distinguishing qualities, collects the hollow core waveguide two-dimensional infrared spectrum information of the samples as the input of the second sensory classification model, performs sensory evaluation and analysis on mainstream cigarette smoke, and compares them with The expert scores are compared to determine the effectiveness of the second sensory classification model.
  • the objective function is adaptively approached to realize the quality evaluation and sensory analysis of mainstream cigarette smoke.
  • Step S6 According to the shallow sensory quality results and the deep sensory quality results, a comprehensive sensory quality result of the mainstream smoke spectrum data of each of the cigarettes is obtained.
  • the step S6 may specifically include:
  • Step S61 Compare the shallow sensory quality results and the deep sensory quality results with expert taste results to obtain the shallow modeling accuracy corresponding to the shallow sensory quality results and the deep sensory quality results. The accuracy of deep modeling corresponding to the sensory quality results.
  • Step S62 Determine the weight of the shallow sensory quality result and the deep sensory quality result according to the shallow modeling accuracy rate and the deep modeling accuracy rate.
  • Step S63 Perform a weighted summation on the shallow sensory quality results and the deep sensory quality results to obtain a comprehensive sensory quality result.
  • the comprehensive sensory quality results are obtained, and the characteristics of the spectral shallow feature network and the deep feature network are organically combined, so that they complement and correct each other, and have outstanding spectral big data Feature extraction capabilities.
  • the sensory evaluation method for cigarette mainstream smoke spectrum data can maximize the spectrum implicit in cigarette mainstream smoke by enhancing the processing of cigarette mainstream smoke spectrum data under limited sample conditions Feature and spatial feature paradigm, thereby effectively reducing the need for training sample size; extracting shallow spectral features can provide guidance information on key components of complex systems for subsequent deep learning, and help improve the extraction accuracy of deep spatial features ; Extract deep spatial features, you can quickly learn effective deep feature representation from training data, and thereby enhance the feature information expression of abnormal samples and normal samples; the present invention extracts spectral features and spatial features separately from shallow to deep, And through the integration of the spectrum-spatial classification framework, the sensory evaluation results of mainstream cigarette smoke are automatically and directly obtained, and accurate screening of unknowns in mainstream smoke is realized.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Chemical & Material Sciences (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Mathematical Physics (AREA)
  • Business, Economics & Management (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Human Resources & Organizations (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Development Economics (AREA)
  • Educational Administration (AREA)
  • Medical Informatics (AREA)
  • Strategic Management (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Economics (AREA)
  • Food Science & Technology (AREA)
  • Combustion & Propulsion (AREA)
  • Medicinal Chemistry (AREA)
  • Databases & Information Systems (AREA)

Abstract

一种卷烟主流烟气光谱数据的感官评价方法，包括：对若干烟支的卷烟主流烟气光谱数据进行增强处理；从卷烟主流烟气光谱数据中提取浅层光谱特征；基于卷烟主流烟气光谱数据和浅层光谱特征，得到各卷烟主流烟气光谱数据的浅层感官品质结果；从卷烟主流烟气光谱数据中提取深层空间特征；基于卷烟主流烟气光谱数据和深层空间特征，得到深层感官品质结果；根据浅层感官品质结果和深层感官品质结果，得到综合感官品质结果。该卷烟主流烟气光谱数据的感官评价方法，由浅入深地分别提取光谱特征和空间特征，并通过融合光谱—空间分类框架，自动直接获得卷烟主流烟气的感官评价结果，实现主流烟气中未知物的准确筛查。

Description

卷烟主流烟气光谱数据的感官评价方法 技术领域
本发明涉及烟草制品质量评价技术领域，尤其涉及一种卷烟主流烟气光谱数据的感官评价方法。
背景技术
卷烟作为一种特殊吸食消费品，其主流烟气中关键组分含量直接影响到卷烟品质与口感，如何有效量化烟气成分与感官评吸之间的关联性以实现产品的精准评价，成为烟草行业亟待解决的关键技术瓶颈。
目前，针对卷烟主流烟气光谱数据的感官评价方法通过专家评吸来得到，即感官评价是通过专业人员对卷烟进行品吸，再对相应的评分项进行打分得到的。这种感官评价方法依赖于人工，而且人工评吸受到感官评吸人员自身因素以及外部环境的影响，这给卷烟评价结果带来了不确定的因素，同时，专家评吸方法存在费时费力、过程繁琐、效率低、感官评价结果不稳定等问题。因此，亟需一种卷烟主流烟气光谱数据的感官评价方法。
发明内容
本发明的目的是提供一种卷烟主流烟气光谱数据的感官评价方法，以解决上述现有技术中的问题，能够提高对卷烟主流烟气的感官评价的效率、准确性和稳定性。
本发明提供了一种卷烟主流烟气光谱数据的感官评价方法，其中，包括：
对若干烟支的卷烟主流烟气光谱数据进行增强处理；
从数据增强处理后的各烟支的所述卷烟主流烟气光谱数据中提取浅层光谱特征；
基于已提取所述浅层光谱特征的各烟支的所述卷烟主流烟气光谱数据和所述浅层光谱特征，得到各所述卷烟主流烟气光谱数据的浅层感官品质结果；
从已提取所述浅层光谱特征的各烟支的所述卷烟主流烟气光谱数据中提取深层空间特征；
基于已提取所述深层空间特征的各烟支的所述卷烟主流烟气光谱数据和所述深层空间特征，得到各所述卷烟主流烟气光谱数据的深层感官品质结果；
根据所述浅层感官品质结果和所述深层感官品质结果，得到各所述卷烟主流烟气光谱数据的综合感官品质结果。
如上所述的卷烟主流烟气光谱数据的感官评价方法，其中，优选的是，所述卷烟主流烟气光谱数据包括中红外光谱数据。
如上所述的卷烟主流烟气光谱数据的感官评价方法，其中，优选的是，所述对若干烟支的卷烟主流烟气光谱数据进行增强处理，具体包括：
对各烟支的卷烟主流烟气光谱数据进行水平翻转处理；
对经过水平翻转处理的各烟支的卷烟主流烟气光谱数据进行随机裁剪处理；
对经过随机裁剪处理的各烟支的卷烟主流烟气光谱数据进行物理扰动处理；
对经过物理扰动处理的各烟支的卷烟主流烟气光谱数据进行组分扰动处理。
如上所述的卷烟主流烟气光谱数据的感官评价方法，其中，优选的是，所述从数据增强处理后的各烟支的所述卷烟主流烟气光谱数据中提取浅层光谱特征，具体包括：
利用光谱向量的Hotelling T 2统计量剔除若干烟支的卷烟主流烟气光谱数据中的离群数据点，以剔除卷烟主流烟气光谱数据中的异常数据；
采用二阶微分、Karl Norris导数滤波器、多元散射校正和小波变换中的至少一种方法对剔除异常数据的各所述卷烟主流烟气光谱数据进行降噪处理。
如上所述的卷烟主流烟气光谱数据的感官评价方法，其中，优选的是，所述基于已提取所述浅层光谱特征的各烟支的所述卷烟主流烟气光谱数据和所述浅层光谱特征，得到各所述卷烟主流烟气光谱数据的浅层感官品质结果，具体包括：
将已提取所述浅层光谱特征的各所述卷烟主流烟气光谱数据输入到预先构建的第一感官分类模型，得到各所述卷烟主流烟气光谱数据的浅层感官品质结果。
如上所述的卷烟主流烟气光谱数据的感官评价方法，其中，优选的是，所述第一感官分类模型是基于主成分分析结合非线性支持向量机构建的，并且所述第一感官分类模型的构建方法具体包括：
基于主成分分析法对降噪处理后的各所述卷烟主流烟气光谱数据进行特征选择，以提取出主流烟气中的各组分在所述卷烟主流烟气光谱数据中的特征峰；
基于非线性支持向量机算法对经过特征峰提取的各所述卷烟主流烟气光谱数据进行训练，得到第一感官分类模型。
如上所述的卷烟主流烟气光谱数据的感官评价方法，其中，优选的是，所述从已提取所述浅层光谱特征的各烟支的所述卷烟主流烟气光谱数据中提取深层空间特征，具体包括：
基于深度残差卷积神经网络，从已提取所述浅层光谱特征的各烟支的所述卷烟主流烟气光谱数据中提取深层空间特征。
如上所述的卷烟主流烟气光谱数据的感官评价方法，其中，优选的是，所述基于已提取所述深层空间特征的各烟支的所述卷烟主流烟气光谱数据和所述深层空间特征，得到各所述卷烟主流烟气光谱数据的深层感官品质结果，具体包括：
将若干个基于所述深度残差卷积神经网络提取的所述深层空间特征，以堆栈的方式输入到支持向量机中，得到第二感官分类模型；
将已提取所述深层空间特征的各所述卷烟主流烟气光谱数据输入到所述第二感官分类模型，得到各所述卷烟主流烟气光谱数据的深层感官品质结果。
如上所述的卷烟主流烟气光谱数据的感官评价方法，其中，优选的是，所述深度残差卷积神经网络的网络参数的确定方法具体包括：
在固定卷烟主流烟气光谱数据集的前提下，以最佳分类误差作为第一目标函数，获取第一最佳网络参数；
以最快运算效率为第二目标函数，获取第二最佳网络参数；
选择与第一最佳网络参数对应的第一最佳卷积核和与所述第二最佳网络参数对应的第二最佳卷积核的卷积核尺寸的平衡点，作为所述深度残差卷积神经网络的最终网络参数。
如上所述的卷烟主流烟气光谱数据的感官评价方法，其中，优选的是，所述根据所述浅层感官品质结果和所述深层感官品质结果，得到各所述卷烟主流烟气光谱数据的综合感官品质结果，具体包括：
分别将所述浅层感官品质结果和所述深层感官品质结果与专家品吸结果进行对比，以得到与所述浅层感官品质结果对应的浅层建模正确率和与所述深层感官品质结果对应的深层建模正确率；
根据所述浅层建模正确率和所述深层建模正确率确定所述浅层感官品质结果和所述深层感官品质结果的权重；
对所述浅层感官品质结果和所述深层感官品质结果进行加权求和，以得到综合感官品质结果。
本发明提供一种卷烟主流烟气光谱数据的感官评价方法，通过对卷烟主流烟气光谱数据进行增强处理，可以在有限的样本条件下，最大限度地提升卷烟主流烟气中隐含的光谱特征与空间特征范式，进而有效降低对训练样本量的需求；提取浅层光谱特征，可以为后续的深度学习提供了复杂体系关键组分的导向性信息，有助于提升深度空间特征的提取精度；提取深层空间特征，可以从训练数据中很快学到有效的深度特征表示，并由此增强异常样本与正常样本的特征信息表达；本发明通过由浅入深地分别提取光谱特征和空间特征，并通过融合光谱—空间分类框架，自动直接获得卷烟主流烟气的感官评价结果，实现主流烟气中未知物的准确筛查。
附图说明
为使本发明的目的、技术方案和优点更加清楚，下面将结合附图对本发明作进一步描述，其中：
图1为本发明提供的卷烟主流烟气光谱数据的感官评价方法的实施例的流程图。
具体实施方式
现在将参照附图来详细描述本公开的各种示例性实施例。对示例性实施例的描述仅仅是说明性的，决不作为对本公开及其应用或使用的任何限制。本公开可以以许多不同的形式实现，不限于这里所述的实施例。提供这些实施例是为了使本公开透彻且完整，并且向本领域技术人员充分表达本公开的范围。应注意到：除非另外具体说明，否则在这些实施例中阐述的部件和步骤的相对布置、材料的组分、数字表达式和数值应被解释为仅仅是示例性的，而不是作为限制。
本公开中使用的“第一”、“第二”：以及类似的词语并不表示任何顺序、数量或者重要性，而只是用来区分不同的部分。“包括”或者“包含”等类似的词语意指在该词前的要素涵盖在该词后列举的要素，并不排除也涵盖其他要素的可能。“上”、“下”等仅用于表示相对位置关系，当被描述对象的绝对位置改变后，则该相对位置关系也可能相应地改变。
在本公开中，当描述到特定部件位于第一部件和第二部件之间时，在该特定部件与第一部件或第二部件之间可以存在居间部件，也可以不存在居间部件。当描述到特定部件连接其它部件时，该特定部件可以与所述其它部件直接连接而不具有居间部件，也可以不与所述其它部件直接连接而具有居间部件。
本公开使用的所有术语(包括技术术语或者科学术语)与本公开所属领域的普通技术人员理解的含义相同，除非另外特别定义。还应当理解，在诸如通用字典中定义的术语应当被解释为具有与它们在相关技术的上下文中的含义相一致的含义，而不应用理想化或极度形式化的意义来解释，除非这里明确地这样定义。
对于相关领域普通技术人员已知的技术、方法和设备可能不作详细讨 论，但在适当情况下，技术、方法和设备应当被视为说明书的一部分。
如图1所示，本实施例提供的卷烟主流烟气光谱数据的感官评价方法在实际执行过程中，具体包括：
步骤S1、对若干烟支的卷烟主流烟气光谱数据进行增强处理。
其中，所述卷烟主流烟气光谱数据包括中红外光谱数据，具体是基于空芯波导二维红外光谱检测技术获得的，空芯波导红外光谱技术作为一种红外增强技术，与常规红外光谱技术相比，空芯光纤技术借助Ag/AgI复合涂层的高反射率，使得入射光源在空芯光纤内形成多次反射而延长光与物质交互作用的光程，能够更高效地提升待测体系的红外吸收强度，从而降低检出限，提高分析的精密度和准确性。
对卷烟主流烟气光谱数据进行增强处理，可以从全局优化的角度降低对训练样本数的需求，可以增强与主流烟气相关组分的光谱(例如为空芯波导二维红外光谱)空间范式，有效降低第一感官分类模型和第二感官分类模型(将在后文描述)的过拟合风险。这样，可以在有限的样本条件下，最大限度地提升卷烟主流烟气中隐含的光谱特征与空间特征范式，推动第二感官分类模型对已有数据的深度挖掘，进而有效降低对训练样本量的需求。
进一步地，在本发明的卷烟主流烟气光谱数据的感官评价方法的一种实施方式中，所述步骤S1具体可以包括：
步骤S11、对各烟支的卷烟主流烟气光谱数据进行水平翻转处理。
步骤S12、对经过水平翻转处理的各烟支的卷烟主流烟气光谱数据进行随机裁剪处理。
考虑到第二感官分类模型中涉及到深度学习的部分对数据量要求最高，即空芯波导二维红外光谱数据中的空间分布特征，该特征可视为图像信息。因此，本发明采用传统的图像增强模式，对其进行水平翻转、随机裁剪等操作，以提升第二感官分类模型对空间图像识别的有效性和鲁棒性。
步骤S13、对经过随机裁剪处理的各烟支的卷烟主流烟气光谱数据进行物理扰动处理。
在实际样本的空芯波导二维红外光谱分析过程中，二维光谱信息不仅 受物质组成信息影响，还受其物理状态的影响。因此，通过改变样本的物理状态，将有效获取更多的空芯波导二维红外光谱范式信息，进而提升深度学习的准确性和有效性。卷烟主流烟气中的物理扰动因素主要包括温度、风速、压力、抽取方式等不同物理状态对卷烟主流烟气的空芯波导二维红外光谱信息的影响。
步骤S14、对经过物理扰动处理的各烟支的卷烟主流烟气光谱数据进行组分扰动处理。
在具体实现中，通过随意添加不同配方的标准样品、改变配方组成等组分扰动方式，可以确定不同组分扰动状态下的卷烟主流烟气空芯波导二维红外光谱的信息变化。
通过以上多种不同的数据增强方式，可以提升实际样品空芯波导二维红外光谱信号的范式信息，进而提升后续的空芯波导二维红外光谱的分类识别性能。
步骤S2、从数据增强处理后的各烟支的所述卷烟主流烟气光谱数据中提取浅层光谱特征。
其中，在本发明的卷烟主流烟气光谱数据的感官评价方法的一种实施方式中，所述步骤S2具体可以包括：
步骤S21、利用光谱向量的Hotelling T 2统计量剔除若干烟支的卷烟主流烟气光谱数据中的离群数据点，以剔除卷烟主流烟气光谱数据中的异常数据。
步骤S22、采用二阶微分、Karl Norris导数滤波器、多元散射校正(multiplicative scatter correction，MSC)和小波变换中的至少一种方法对剔除异常数据的各所述卷烟主流烟气光谱数据进行降噪处理。
通过降噪处理，可以降低噪声干扰，使卷烟主流烟气光谱数据中的特征峰更加明显，便于从卷烟主流烟气光谱数据中的背景中提取特征峰，以此来提高信噪比。而且，通过数据筛选和降噪处理，便于引导后续的光谱解析方法准确提取待测物质的数据特征。
步骤S3、基于已提取所述浅层光谱特征的各烟支的所述卷烟主流烟气光谱数据和所述浅层光谱特征，得到各所述卷烟主流烟气光谱数据的浅层 感官品质结果。
提取浅层光谱特征的目的是从复杂、变动的光谱(例如为空芯波导二维红外光谱)信号中提取样品关键组分的光谱特征，并实现光谱空间降维。具体地，将已提取所述浅层光谱特征的各所述卷烟主流烟气光谱数据输入到预先构建的第一感官分类模型，得到各所述卷烟主流烟气光谱数据的浅层感官品质结果。通过浅层光谱特征的提取，可以为后续的深度学习提供了复杂体系关键组分的导向性信息，有助于提升深度空间特征的提取精度。
进一步地，所述第一感官分类模型是基于主成分分析(Principal Component Analysis，PCA)结合非线性支持向量机(Support vector machine，SVM)构建的，并且所述第一感官分类模型的构建方法具体包括：
基于主成分分析法对降噪处理后的各所述卷烟主流烟气光谱数据进行特征选择，以提取出主流烟气中的各组分在所述卷烟主流烟气光谱数据中的特征峰；
基于非线性支持向量机算法对经过特征峰提取的各所述卷烟主流烟气光谱数据进行训练，得到第一感官分类模型。
与线性支持向量机相比，非线性支持向量机的分类过程较模糊，只要输入提取出来的特征峰，就能自己建立判别关系和分类关系。
进一步地，所述第一感官分类模型的分类结果至少包括好、中和差。在本发明中，第一感官分类模型的分类结果为好、中和差，需要说明的是，本发明对判第一感官分类模型的分类结果和数量不作具体限定，通过定义参数和变更权重，可以得到其他分类结果。
更进一步地，所述第一感官分类模型在训练过程中，利用专家感官评吸得分对所述第一感官分类模型输出的第一分类判断值进行监督，这样可以实现第一感官分类模型的验证和更新。
在本发明的卷烟主流烟气光谱数据的感官评价方法的一种实施方式中，所述第一感官分类模型通过如下训练方法进行训练：
首先，将所述卷烟主流烟气光谱数据的训练集输入到所述第一感官分类模型模型。具体可以包括：将所述卷烟主流烟气光谱数据的原始训练集中的异常数据进行剔除；对剔除异常数据的原始训练集中的各所述卷烟主 流烟气光谱数据进行降噪处理；将降噪处理后的原始训练集中的各所述卷烟主流烟气光谱数据输入到所述第一感官分类模型。
然后，根据所述第一分类判断值和所述专家感官评吸得分，得到第一目标函数，将所述第一目标函数的梯度反传至所述所述第一感官分类模型模型。
最后，当基于所述第一分类判断值和所述专家感官评吸得分得到的所述第一目标函数的函数值到达设定值时，停止训练。
步骤S4、从已提取所述浅层光谱特征的各烟支的所述卷烟主流烟气光谱数据中提取深层空间特征。
具体地，基于深度残差卷积神经网络，从已提取所述浅层光谱特征的各烟支的所述卷烟主流烟气光谱数据中提取深层空间特征。当复杂体系中出现未知干扰物时，原有关键组分的空间拓扑信息将在一定程度内造成畸变，而深度残差卷积神经网络可以针对这一变化，从训练数据中很快学到有效的深度特征表示，并由此增强异常样本与正常样本的特征信息表达。
步骤S5、基于已提取所述深层空间特征的各烟支的所述卷烟主流烟气光谱数据和所述深层空间特征，得到各所述卷烟主流烟气光谱数据的深层感官品质结果。
提取深层空间特征的目的，是借助于深度残差卷积神经网络的方法，以平移不变的方式完成深层空间特征的提取及增强。其中，所述深度残差卷积神经网络的网络参数的确定方法具体包括：
在固定卷烟主流烟气光谱数据集的前提下，以最佳分类误差作为第一目标函数，获取第一最佳网络参数；
以最快运算效率为第二目标函数，获取第二最佳网络参数；
选择与第一最佳网络参数对应的第一最佳卷积核和与所述第二最佳网络参数对应的第二最佳卷积核的卷积核尺寸的平衡点，作为所述深度残差卷积神经网络的最终网络参数。
进一步地，在本发明的卷烟主流烟气光谱数据的感官评价方法的一种实施方式中，所述步骤S5具体可以包括：
步骤S51、将若干个基于所述深度残差卷积神经网络提取的所述深层空 间特征，以堆栈的方式输入到支持向量机中，得到第二感官分类模型。
进一步地，所述第二感官分类模型的分类结果为感官得分。在本发明中，第二感官分类模型的分类结果为感官得分，需要说明的是，本发明对判第二感官分类模型的分类结果和数量不作具体限定，通过定义参数和变更权重，可以得到其他分类结果。
步骤S52、将已提取所述深层空间特征的各所述卷烟主流烟气光谱数据输入到所述第二感官分类模型，得到各所述卷烟主流烟气光谱数据的深层感官品质结果。
本发明引入深度学习策略以模拟不同专家的感官评吸过程。本发明在具体实现中，在深度学习过程中，将通过主流烟气分析单只烟的感官数据这一目标问题分成配置了不同权重和参数的预测模型的多分类问题，以模拟不同专家之间的感官差异，并将不同卷烟样品的主流烟气成分视为基于专家先验知识的多标记问题，利用人工专家评吸得到的专家标签的得分信息改进深度学习模型的监督信息。在标记的改善过程中，利用通过第二感官分类模型得到的得分信息与通过人工专家评吸得到的专家标签的得分信息的共生关系，来改善已经得到的深度学习训练结果。
进一步地，本发明还选择具有不同区分度的品质的标准卷烟样品，采集样品的空芯波导二维红外光谱信息作为第二感官分类模型的输入，对卷烟主流烟气进行感官评价分析，并与专家评分进行比对，以确定第二感官分类模型的有效性。这样，通过反复迭代、升级，可以在难以建立精确评估模型的情况下，通过分析空芯波导二维红外光谱数据特征及其内在规律，以在于感官评价相关的特征段进行局部动态分析的方式，自适应地逼近目标函数，进而实现卷烟主流烟气的品质评价与感官分析。
步骤S6、根据所述浅层感官品质结果和所述深层感官品质结果，得到各所述卷烟主流烟气光谱数据的综合感官品质结果。
其中，在本发明的卷烟主流烟气光谱数据的感官评价方法的一种实施方式中，所述步骤S6具体可以包括：
步骤S61、分别将所述浅层感官品质结果和所述深层感官品质结果与专家品吸结果进行对比，以得到与所述浅层感官品质结果对应的浅层建模正 确率和与所述深层感官品质结果对应的深层建模正确率。
步骤S62、根据所述浅层建模正确率和所述深层建模正确率确定所述浅层感官品质结果和所述深层感官品质结果的权重。
步骤S63、对所述浅层感官品质结果和所述深层感官品质结果进行加权求和，以得到综合感官品质结果。
根据浅层感官品质结果和深层感官品质结果得到综合感官品质结果，有机结合了光谱浅层特征网络与深层特征网络之间的特点，使其互为补充、互为校正，具备突出的光谱大数据特征提取能力。
本发明实施例提供的卷烟主流烟气光谱数据的感官评价方法，通过对卷烟主流烟气光谱数据进行增强处理，可以在有限的样本条件下，最大限度地提升卷烟主流烟气中隐含的光谱特征与空间特征范式，进而有效降低对训练样本量的需求；提取浅层光谱特征，可以为后续的深度学习提供了复杂体系关键组分的导向性信息，有助于提升深度空间特征的提取精度；提取深层空间特征，可以从训练数据中很快学到有效的深度特征表示，并由此增强异常样本与正常样本的特征信息表达；本发明通过由浅入深地分别提取光谱特征和空间特征，并通过融合光谱—空间分类框架，自动直接获得卷烟主流烟气的感官评价结果，实现主流烟气中未知物的准确筛查。
至此，已经详细描述了本公开的各实施例。为了避免遮蔽本公开的构思，没有描述本领域所公知的一些细节。本领域技术人员根据上面的描述，完全可以明白如何实施这里公开的技术方案。
虽然已经通过示例对本公开的一些特定实施例进行了详细说明，但是本领域的技术人员应该理解，以上示例仅是为了进行说明，而不是为了限制本公开的范围。本领域的技术人员应该理解，可在不脱离本公开的范围和精神的情况下，对以上实施例进行修改或者对部分技术特征进行等同替换。本公开的范围由所附权利要求来限定。

Claims (10)

  1. 一种卷烟主流烟气光谱数据的感官评价方法，其特征在于，包括：
    对若干烟支的卷烟主流烟气光谱数据进行增强处理；
    从数据增强处理后的各烟支的所述卷烟主流烟气光谱数据中提取浅层光谱特征；
    基于已提取所述浅层光谱特征的各烟支的所述卷烟主流烟气光谱数据和所述浅层光谱特征，得到各所述卷烟主流烟气光谱数据的浅层感官品质结果；
    从已提取所述浅层光谱特征的各烟支的所述卷烟主流烟气光谱数据中提取深层空间特征；
    基于已提取所述深层空间特征的各烟支的所述卷烟主流烟气光谱数据和所述深层空间特征，得到各所述卷烟主流烟气光谱数据的深层感官品质结果；
    根据所述浅层感官品质结果和所述深层感官品质结果，得到各所述卷烟主流烟气光谱数据的综合感官品质结果。
  2. 根据权利要求1所述的卷烟主流烟气光谱数据的感官评价方法，其特征在于，所述卷烟主流烟气光谱数据包括中红外光谱数据。
  3. 根据权利要求1所述的卷烟主流烟气光谱数据的感官评价方法，其特征在于，所述对若干烟支的卷烟主流烟气光谱数据进行增强处理，具体包括：
    对各烟支的卷烟主流烟气光谱数据进行水平翻转处理；
    对经过水平翻转处理的各烟支的卷烟主流烟气光谱数据进行随机裁剪处理；
    对经过随机裁剪处理的各烟支的卷烟主流烟气光谱数据进行物理扰动处理；
    对经过物理扰动处理的各烟支的卷烟主流烟气光谱数据进行组分扰动处理。
  4. 根据权利要求1所述的卷烟主流烟气光谱数据的感官评价方法，其特征在于，所述从数据增强处理后的各烟支的所述卷烟主流烟气光谱数据 中提取浅层光谱特征，具体包括：
    利用光谱向量的Hotelling T 2统计量剔除若干烟支的卷烟主流烟气光谱数据中的离群数据点，以剔除卷烟主流烟气光谱数据中的异常数据；
    采用二阶微分、Karl Norris导数滤波器、多元散射校正和小波变换中的至少一种方法对剔除异常数据的各所述卷烟主流烟气光谱数据进行降噪处理。
  5. 根据权利要求4所述的卷烟主流烟气光谱数据的感官评价方法，其特征在于，所述基于已提取所述浅层光谱特征的各烟支的所述卷烟主流烟气光谱数据和所述浅层光谱特征，得到各所述卷烟主流烟气光谱数据的浅层感官品质结果，具体包括：
    将已提取所述浅层光谱特征的各所述卷烟主流烟气光谱数据输入到预先构建的第一感官分类模型，得到各所述卷烟主流烟气光谱数据的浅层感官品质结果。
  6. 根据权利要求5所述的卷烟主流烟气光谱数据的感官评价方法，其特征在于，所述第一感官分类模型是基于主成分分析结合非线性支持向量机构建的，并且所述第一感官分类模型的构建方法具体包括：
    基于主成分分析法对降噪处理后的各所述卷烟主流烟气光谱数据进行特征选择，以提取出主流烟气中的各组分在所述卷烟主流烟气光谱数据中的特征峰；
    基于非线性支持向量机算法对经过特征峰提取的各所述卷烟主流烟气光谱数据进行训练，得到第一感官分类模型。
  7. 根据权利要求1所述的卷烟主流烟气光谱数据的感官评价方法，其特征在于，所述从已提取所述浅层光谱特征的各烟支的所述卷烟主流烟气光谱数据中提取深层空间特征，具体包括：
    基于深度残差卷积神经网络，从已提取所述浅层光谱特征的各烟支的所述卷烟主流烟气光谱数据中提取深层空间特征。
  8. 根据权利要求7所述的卷烟主流烟气光谱数据的感官评价方法，其特征在于，所述基于已提取所述深层空间特征的各烟支的所述卷烟主流烟气光谱数据和所述深层空间特征，得到各所述卷烟主流烟气光谱数据的深 层感官品质结果，具体包括：
    将若干个基于所述深度残差卷积神经网络提取的所述深层空间特征，以堆栈的方式输入到支持向量机中，得到第二感官分类模型；
    将已提取所述深层空间特征的各所述卷烟主流烟气光谱数据输入到所述第二感官分类模型，得到各所述卷烟主流烟气光谱数据的深层感官品质结果。
  9. 根据权利要求7所述的卷烟主流烟气光谱数据的感官评价方法，其特征在于，所述深度残差卷积神经网络的网络参数的确定方法具体包括：
    在固定卷烟主流烟气光谱数据集的前提下，以最佳分类误差作为第一目标函数，获取第一最佳网络参数；
    以最快运算效率为第二目标函数，获取第二最佳网络参数；
    选择与第一最佳网络参数对应的第一最佳卷积核和与所述第二最佳网络参数对应的第二最佳卷积核的卷积核尺寸的平衡点，作为所述深度残差卷积神经网络的最终网络参数。
  10. 根据权利要求1所述的卷烟主流烟气光谱数据的感官评价方法，其特征在于，所述根据所述浅层感官品质结果和所述深层感官品质结果，得到各所述卷烟主流烟气光谱数据的综合感官品质结果，具体包括：
    分别将所述浅层感官品质结果和所述深层感官品质结果与专家品吸结果进行对比，以得到与所述浅层感官品质结果对应的浅层建模正确率和与所述深层感官品质结果对应的深层建模正确率；
    根据所述浅层建模正确率和所述深层建模正确率确定所述浅层感官品质结果和所述深层感官品质结果的权重；
    对所述浅层感官品质结果和所述深层感官品质结果进行加权求和，以得到综合感官品质结果。
PCT/CN2021/079141 2021-01-14 2021-03-04 卷烟主流烟气光谱数据的感官评价方法 WO2021253874A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
EP21806940.9A EP3961189B1 (en) 2021-01-14 2021-03-04 Cigarette mainstream smoke spectral data sensory evaluation method
JP2021568873A JP7203252B2 (ja) 2021-01-14 2021-03-04 紙巻きタバコの主流煙のスペクトルデータの官能評価方法
US17/595,852 US11525774B2 (en) 2021-01-14 2021-03-04 Sensory evaluation method for spectral data of mainstream smoke

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110050500.6 2021-01-14
CN202110050500.6A CN112749906A (zh) 2021-01-14 2021-01-14 卷烟主流烟气光谱数据的感官评价方法

Publications (1)

Publication Number Publication Date
WO2021253874A1 true WO2021253874A1 (zh) 2021-12-23

Family

ID=75652042

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/079141 WO2021253874A1 (zh) 2021-01-14 2021-03-04 卷烟主流烟气光谱数据的感官评价方法

Country Status (5)

Country Link
US (1) US11525774B2 (zh)
EP (1) EP3961189B1 (zh)
JP (1) JP7203252B2 (zh)
CN (1) CN112749906A (zh)
WO (1) WO2021253874A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116189803B (zh) * 2023-03-31 2024-02-02 江苏中烟工业有限责任公司 一种卷烟叶组配方的稳质增效方法及其应用

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH09304371A (ja) * 1996-05-09 1997-11-28 Mitsubishi Rayon Co Ltd たばこ煙の分別評価装置及びそれを用いたたばこ煙の分別評価方法
CN1975705A (zh) * 2005-11-28 2007-06-06 颐中烟草(集团)有限公司 基于回归函数估计svm的卷烟内在质量指标评估方法
CN107014768A (zh) * 2017-05-26 2017-08-04 云南中烟工业有限责任公司 卷烟主流烟气整体质量特征的红外光谱分析方法
CN107646089A (zh) * 2015-03-06 2018-01-30 英国质谱公司 光谱分析
CN108414471A (zh) * 2018-01-10 2018-08-17 浙江中烟工业有限责任公司 一种基于近红外光谱与感官评吸互信息判别感官表征信息的方法
CN108801968A (zh) * 2018-06-05 2018-11-13 天津大学 一种基于真空红外光谱的烟丝整体质量特征分析方法
CN109444326A (zh) * 2019-01-10 2019-03-08 云南中烟工业有限责任公司 一种卷烟主流烟气主成分捕集检测装置及其分析方法
CN110132880A (zh) * 2019-06-18 2019-08-16 浙江中烟工业有限责任公司 一种基于近红外光谱的烟叶整体感官质量评价方法

Family Cites Families (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6744503B2 (en) * 2001-12-20 2004-06-01 Philip Morris Incorporated Monitoring of vapor phase polycyclic aromatic hydrocarbons
CN1525394A (zh) 2003-02-25 2004-09-01 颐中烟草(集团)有限公司 卷烟感官评吸和烟气指标的神经网络预测方法
CN103091280A (zh) * 2013-01-14 2013-05-08 安徽中烟工业有限责任公司 一种卷烟抽吸轻松度的测定方法
GB201315492D0 (en) * 2013-08-30 2013-10-16 British American Tobacco Co Smoking Article
CN103674874B (zh) * 2013-11-25 2017-10-10 广东中烟工业有限责任公司 一种通过紫外光谱评价烟叶醇化质量的方法
CN104198573B (zh) 2014-09-20 2017-04-05 国家烟草质量监督检验中心 一种卷烟主流烟气在线分析方法及其专用的在线分析装置
CN104568825B (zh) * 2015-01-21 2017-03-01 四川中烟工业有限责任公司 烤烟烟叶打叶复烤模块组配的方法
GB201611596D0 (en) 2016-07-04 2016-08-17 British American Tobacco Investments Ltd Apparatus and method for classifying a tobacco sample into one of a predefined set of taste categories
CN107179379B (zh) 2017-06-22 2019-08-02 河北中烟工业有限责任公司 一种卷烟主流烟气成份感官风格指标的评估方法
WO2019085369A1 (zh) * 2017-10-31 2019-05-09 高大启 一种电子鼻仪器和烟草与烟草制品感官质量评价方法
CN108764142A (zh) * 2018-05-25 2018-11-06 北京工业大学 基于3dcnn的无人机影像森林烟雾检测和分类方法
CN209525317U (zh) * 2019-01-10 2019-10-22 云南中烟工业有限责任公司 一种卷烟主流烟气主成分捕集检测装置
CN111652039B (zh) * 2020-04-13 2023-04-18 上海海洋大学 基于残差网络和特征融合模块的高光谱遥感地物分类方法

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH09304371A (ja) * 1996-05-09 1997-11-28 Mitsubishi Rayon Co Ltd たばこ煙の分別評価装置及びそれを用いたたばこ煙の分別評価方法
CN1975705A (zh) * 2005-11-28 2007-06-06 颐中烟草(集团)有限公司 基于回归函数估计svm的卷烟内在质量指标评估方法
CN107646089A (zh) * 2015-03-06 2018-01-30 英国质谱公司 光谱分析
CN107014768A (zh) * 2017-05-26 2017-08-04 云南中烟工业有限责任公司 卷烟主流烟气整体质量特征的红外光谱分析方法
CN108414471A (zh) * 2018-01-10 2018-08-17 浙江中烟工业有限责任公司 一种基于近红外光谱与感官评吸互信息判别感官表征信息的方法
CN108801968A (zh) * 2018-06-05 2018-11-13 天津大学 一种基于真空红外光谱的烟丝整体质量特征分析方法
CN109444326A (zh) * 2019-01-10 2019-03-08 云南中烟工业有限责任公司 一种卷烟主流烟气主成分捕集检测装置及其分析方法
CN110132880A (zh) * 2019-06-18 2019-08-16 浙江中烟工业有限责任公司 一种基于近红外光谱的烟叶整体感官质量评价方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP3961189A4 *

Also Published As

Publication number Publication date
JP2022538731A (ja) 2022-09-06
EP3961189A1 (en) 2022-03-02
CN112749906A (zh) 2021-05-04
EP3961189A4 (en) 2022-07-27
JP7203252B2 (ja) 2023-01-12
EP3961189B1 (en) 2023-06-14
US11525774B2 (en) 2022-12-13
US20220268696A1 (en) 2022-08-25

Similar Documents

Publication Publication Date Title
CN114444986B (zh) 产品分析方法及系统及装置及介质
CN109632693A (zh) 一种基于blstm-rnn的太赫兹光谱识别方法
CN107655850A (zh) 基于近红外光谱的非线性建模方法及系统
CN107679569A (zh) 基于自适应超图算法的拉曼光谱物质自动识别方法
Zhou et al. A monitoring system for the segmentation and grading of broccoli head based on deep learning and neural networks
CN113030001B (zh) 一种水果糖度检测方法及系统
CN108763096A (zh) 基于深度信念网络算法支持向量机的软件缺陷预测方法
CN108802002B (zh) 一种快速无损鉴别解除滞育的蚕卵拉曼光谱模型构建方法
Yang et al. Determination of the soluble solids content in korla fragrant pears based on visible and near-infrared spectroscopy combined with model analysis and variable selection
WO2021253874A1 (zh) 卷烟主流烟气光谱数据的感官评价方法
CN106446124A (zh) 一种基于网络关系图的网站分类方法
WO2023108514A1 (zh) 茶叶近红外光谱分析中谱峰自动检测与重构方法及系统
Xu et al. Nondestructive detection of total soluble solids in grapes using VMD‐RC and hyperspectral imaging
Sun Wood quality defect detection based on deep learning and multicriteria framework
Shao et al. A new approach to discriminate varieties of tobacco using vis/near infrared spectra
CN113903407A (zh) 成分识别方法、装置、电子设备和存储介质
WO2022151570A1 (zh) 卷烟主流烟气的品质评价方法
CN110186871A (zh) 一种茶鲜叶产地的判别方法
CN114091794A (zh) 专利价值评估模型训练方法、评估方法、装置及其设备
RU2797495C1 (ru) Способ использования спектроскопических данных для определения органолептических характеристик сигаретного дыма основного потока
CN104897608A (zh) 一种基于近红外光谱技术的乌龙茶品质鉴定方法
CN113138181B (zh) 一种对清香型原酒品质分级的方法
CN111220565B (zh) 一种基于cpls的红外光谱测量仪器标定迁移方法
Huang et al. Neural network for classification of Chinese zither panel wood via near-infrared spectroscopy
Zhang et al. Nondestructive detection of moldy core in apples based on one-dimensional convolutional Neural Networks

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2021568873

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 2021806940

Country of ref document: EP

Effective date: 20211126

NENP Non-entry into the national phase

Ref country code: DE