CN104809230A - Cigarette sensory quality evaluation method based on multi-classifier integration - Google Patents
Cigarette sensory quality evaluation method based on multi-classifier integration Download PDFInfo
- Publication number
- CN104809230A CN104809230A CN201510232615.1A CN201510232615A CN104809230A CN 104809230 A CN104809230 A CN 104809230A CN 201510232615 A CN201510232615 A CN 201510232615A CN 104809230 A CN104809230 A CN 104809230A
- Authority
- CN
- China
- Prior art keywords
- cigarette
- sensory quality
- sensory
- data
- assessment
- 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.)
- Pending
Links
Abstract
The invention provides a cigarette sensory quality evaluation method based on multi-classifier integration. The method comprises the following steps: acquiring finished cigarette sensory quality to-be-evaluated data, namely, cigarette chemical component indexes; performing normalization processing on the acquired finished cigarette sensory quality to-be-evaluated data based on finished cigarette sensory evaluation history data; performing cigarette sensory quality evaluation by using a cigarette sensory quality evaluation model based on multi-classifier integration to obtain a cigarette sensory quality evaluation result. The cigarette sensory quality evaluation method based on multi-classifier integration is applied to the product design and evaluation processes of finished cigarettes in cigarette enterprises. When cigarette experts perform sensory quality evaluation, a plurality of practically-measured chemical component indexes are given specific to a certain cigarette finished product, so that the sensory quality evaluation result of the product can be obtained at high accuracy, and decisions are made instead of or in a way of assisting the experts in product production and design processes.
Description
Technical field
The present invention relates to infotech and technical field of automation, be specifically related to a kind of Sensory Quality of Cigarette appraisal procedure based on Multi-classifers integrated.
Background technology
In production of cigarettes process, be difficult to set up effective mathematical model for the complex relationship of the physical and chemical index of tobacco and the aesthetic quality of cigarette, therefore, in the new product development and product maintenance process of tobacco and goods thereof, the artificial sense mainly through product cigarette expert is smoked panel test and to be evaluated cigarette product aesthetic quality index.Obviously, this mode of production of manually repeatedly smokeing panel test of relying on completely can the efficiency of greatly effect appraise result, cannot meet the requirement of enterprise to production rapidity.
In order to solve, subjectivity in the process of smokeing panel test is strong, the problem of inefficiency, scholars bring into use the method for data mining to evaluate Sensory Quality of Cigarette, make every effort to the mapping ruler going out physical and chemical index and aesthetic quality from a large amount of tobacco extracting data, complete the sense organ prediction and evaluation to cigarette with auxiliary or substitute cigarette expert.The current main intelligent sensory evaluation's problem solving finished cigarettes with BP neural net method or support vector machine method.
But, for the sensory evaluation's historical data existed with the form of continuous data, the means that existing research uses matching to predict mostly carry out Sensory Quality of Cigarette assessment, and for the process of smokeing panel test of product cigarette expert, each expert gives a mark to Sensory quality index for segmentation with 0.5, the continuous data that experiment uses is actual is the mean value of several expert analysis mode, and namely each expert meets the assessment result of enterprise requirements actual is discrete.Therefore, we can be considered as a classification forecasting problem cigarette sensory evaluation problem.Compared with the matching predicting means used in studying with current great majority, the method for classification conforms with the actual mass requirement of enterprise more, and for this abstract index of aesthetic quality, the fitting result pursuing successive value also can limit the production elasticity of enterprise; Meanwhile, at Data Mining, the method for classification is abundanter than the method for matching, uses sorting technique also can have better theoretical foundation to a certain extent.
In addition, finished cigarettes data have small sample, high dimension, very noisy, nonlinear feature.Except using matching predicting means, the applied research in current cigarette sensory evaluation field is the same with traditional mode recognition methods, still puts forth effort on and uses Individual forecast model and improve forecast model.
Summary of the invention
For prior art Problems existing, the invention provides a kind of Sensory Quality of Cigarette appraisal procedure based on Multi-classifers integrated.
Technical scheme of the present invention is:
Based on a Sensory Quality of Cigarette appraisal procedure for Multi-classifers integrated, comprise the steps:
Step 1: gather tobacco product sense organ data to be assessed, i.e. cigarette chemical composition index;
Cigarette chemical composition index comprises: total reducing sugar amount, reducing sugar, nicotine amount, total volatile alkaline, nitrogen pool, nicotine nitrogen, protein, schmuck value, nitrogen base ratio, chlorinty, potassium content, sugared alkali ratio, ammonia state alkali;
Step 2: the tobacco product sense organ data to be assessed gathered are normalized based on tobacco product sensory evaluation historical data;
Step 3: utilize the Sensory Quality of Cigarette assessment models based on Multi-classifers integrated to carry out Sensory Quality of Cigarette assessment;
The described Sensory Quality of Cigarette assessment models based on Multi-classifers integrated is: according to the historical data of tobacco product sensory evaluation, utilize the different sorting techniques in Data Mining to build the single sorter describing relation between cigarette chemical composition index and sensory evaluating smoking's index classification result respectively, and utilize the Multi-classifers integrated model that described single sorter builds;
Described sensory evaluating smoking's index comprises: gloss, fragrance, harmony, assorted gas, stimulation, pleasant impression;
Step 4: obtain Sensory Quality of Cigarette assessment result.
The Sensory Quality of Cigarette assessment models based on Multi-classifers integrated described in step 3 is set up as follows:
Step 3-1: the historical data gathering tobacco product sensory evaluation, sets up Sensory Quality of Cigarette assessment training data sample set;
Sensory Quality of Cigarette assessment training data sample set comprises the expert analysis mode result of cigarette chemical composition index and sensory evaluating smoking's index;
Step 3-2: pre-service is carried out to Sensory Quality of Cigarette assessment training data sample set: according to the specific features of aesthetic quality's judgment criteria and data, respectively discretize is carried out to the expert analysis mode result of 6 kinds of sensory evaluating smoking's indexs; Cigarette chemical composition index is normalized simultaneously;
Step 3-3: utilize the different sorting techniques in pretreated data and Data Mining, builds the classification mathematical prediction model being used for describing relation between cigarette chemical composition and sensory evaluating smoking's index classification result, i.e. single sorter;
Different sorting techniques in described Data Mining, comprising: decision tree C4.5 method, BP neural net method, k-near neighbor method and support vector machine method, and wherein k-near neighbor method gets two different k values, obtain two different single sorters;
Step 3-4: utilize each single sorter to classify to Sensory Quality of Cigarette assessment training data sample set, using the weight of the classification results accuracy rate of each single sorter as each single sorter, build the Sensory Quality of Cigarette assessment models based on Multi-classifers integrated.
The expert analysis mode result of the sensory evaluating smoking's index in described Sensory Quality of Cigarette assessment data sample set obtains by averaging after multiple expert estimation.
Sensory Quality of Cigarette assessment result is obtained described in step 4, specifically: the tobacco product sense organ data to be assessed for current collection calculate sensory evaluating smoking's index classification result of each single sorter respectively, then the weight of single sorter corresponding for each classification is sued for peace, using the classification of classification maximum for weight sum as sensory evaluating smoking's index, i.e. Sensory Quality of Cigarette assessment result.
Beneficial effect
Sensory Quality of Cigarette appraisal procedure based on Multi-classifers integrated of the present invention, is applied to tobacco enterprise in the product design of tobacco product and evaluation process.Cigarette expert carries out aesthetic quality when assessing, based on the Sensory Quality of Cigarette appraisal procedure based on Multi-classifers integrated given by the present invention, for a certain cigarette finished product, the some indexs of chemical composition that its actual measurement given obtains, can obtain aesthetic quality's assessment result of this product in higher precision, replacement or auxiliary expert carry out decision-making in production design process.
Institute of the present invention extracting method can carry out aesthetic quality's evaluation prediction with sensory evaluating smoking's index of higher accuracy to tobacco product, replacement or supplement cigarette expert carry out the sensory evaluation of finished cigarettes to a certain extent, efficiently cigarette product design, production and maintenance are carried out for tobacco enterprise scien`, avoid the duplication of labour, increase work efficiency and there is realistic meaning; The inventive method can be different from existing finished cigarettes method for evaluating quality, the value of the quality evaluation historical data of abundant digging utilization cigarette, compensate for the shortcoming of single sorter model with effective categorizer integration method, for tobacco, enterprise has practical value.
Accompanying drawing explanation
Fig. 1 is the Sensory Quality of Cigarette appraisal procedure process flow diagram based on Multi-classifers integrated of the specific embodiment of the invention;
Fig. 2 is the Sensory Quality of Cigarette assessment models Establishing process figure based on Multi-classifers integrated of the specific embodiment of the invention;
Fig. 3 is the Sensory Quality of Cigarette assessment models structural representation based on Multi-classifers integrated of the specific embodiment of the invention;
Fig. 4 is five kinds of sorting techniques and the inventive method comparison diagram of the specific embodiment of the invention.
Embodiment
Below in conjunction with accompanying drawing, the specific embodiment of the present invention is elaborated.
According to technological requirement and the quality requirements of tobacco product sensory evaluation, present embodiment inputs as data based on these 13 kinds of chemical composition indexs of total reducing sugar amount, reducing sugar, nicotine amount, total volatile alkaline, nitrogen pool, nicotine nitrogen, protein, schmuck value, nitrogen base ratio, chlorinty, potassium content, sugared alkali ratio and ammonia state alkali; Set up the Sensory Quality of Cigarette assessment models based on Multi-classifers integrated based on sensory evaluating smoking's index in gloss, fragrance, harmony, assorted gas, stimulation, pleasant impression 6 as data output and carried out experimental verification.
The Sensory Quality of Cigarette appraisal procedure based on Multi-classifers integrated of present embodiment, as shown in Figure 1, comprises the steps:
Step 1: gather tobacco product sense organ data to be assessed, i.e. cigarette chemical composition index;
Cigarette chemical composition index comprises: total reducing sugar amount, reducing sugar, nicotine amount, total volatile alkaline, nitrogen pool, nicotine nitrogen, protein, schmuck value, nitrogen base ratio, chlorinty, potassium content, sugared alkali ratio, ammonia state alkali;
Step 2: the tobacco product sense organ data to be assessed gathered are normalized based on tobacco product sensory evaluation historical data;
Step 3: utilize the Sensory Quality of Cigarette assessment models based on Multi-classifers integrated to carry out Sensory Quality of Cigarette assessment;
The described Sensory Quality of Cigarette assessment models based on Multi-classifers integrated is: according to the historical data of tobacco product sensory evaluation, utilize the different sorting techniques in Data Mining to build the single sorter describing relation between cigarette chemical composition index and sensory evaluating smoking's index classification result respectively, and utilize the Multi-classifers integrated model that described single sorter builds;
Described sensory evaluating smoking's index comprises: gloss, fragrance, harmony, assorted gas, stimulation, pleasant impression;
The described Sensory Quality of Cigarette assessment models based on Multi-classifers integrated is set up as follows:
Step 3-1: the historical data gathering tobacco product sensory evaluation, sets up Sensory Quality of Cigarette assessment training data sample set;
Sensory Quality of Cigarette assessment training data sample set comprises the expert analysis mode result of cigarette chemical composition index and sensory evaluating smoking's index;
The expert analysis mode result of the sensory evaluating smoking's index in described Sensory Quality of Cigarette assessment data sample set obtains by averaging after multiple expert estimation.
The historical data of collecting tobacco product sensory evaluation assess from the cigarette expert artificial sense of tobacco enterprise assesses training data sample set to set up the Sensory Quality of Cigarette carrying out classifying prediction, historical data is arranged, delete that some repeat or there are the data lacked, last each group historical data obtained includes 13 kinds of chemical composition indexs and 6 kinds of sensory evaluating smoking's indexs, the Sensory Quality of Cigarette assessment training data sample set of present embodiment comprises certain tobacco enterprise tobacco product data of 2010 ~ 2,012 three, amounts to 684 groups of data.
Data all obtain by averaging after multiple expert estimation, the decision attribute that each index uses respectively as classification, carry out six groups of different experiments according to six groups of disparity items.Aesthetic quality's judgment criteria of expert estimation foundation is as shown in table 1.
Table 1 Sensory Quality of Cigarette judgment criteria
Step 3-2: pre-service is carried out to Sensory Quality of Cigarette assessment training data sample set: according to the specific features of aesthetic quality's judgment criteria and data, respectively discretize is carried out to the expert analysis mode result of 6 kinds of sensory evaluating smoking's indexs; Cigarette chemical composition index is normalized simultaneously;
Step 3-2-1: respectively discretize is carried out to the expert analysis mode result of 6 kinds of sensory evaluating smoking's indexs;
Discretization method is as shown in table 2, and wherein discrete segment length equals 0.5.For gloss, marking judgment criteria is 3-5, and there are not the data being less than 3.75 in real data, therefore [3.75,4.25] is classified as the 1st class, (4.25,4.75] be classified as the 2nd class, (4.75,5.25] be classified as the 3rd class.
Table 2 expert analysis mode result discrete method
Step 3-2-2: the normalization of cigarette chemical composition index;
Definition set X={x
1, x
2..., x
684be the value of conditional attribute, max (X) is the maximum occurrences in this set, and min (X) is the minimum value in this set, then to x
n, n ∈ 1,2 ..., 684} has the result x ' after normalization
n;
Step 3-3: utilize the different sorting techniques in pretreated data and Data Mining, builds the classification mathematical prediction model being used for describing relation between cigarette chemical composition and sensory evaluating smoking's index classification result, i.e. single sorter;
Different sorting techniques in described Data Mining, comprising: decision tree C4.5 method, BP neural net method, k-near neighbor method and support vector machine method, and wherein k-near neighbor method gets two different k values, obtain two different single sorters;
Step 3-3-1: set up tobacco product sensory evaluating smoking index prediction model, namely based on the single sorter of k-near neighbor method based on k-near neighbor method;
Parameter k value in k-near neighbor method represents that the sample choosing k Euclidean distance nearer carries out simple vote, and k value difference can cause larger classifying quality difference, and present embodiment is chosen k=5 and k=3 two kinds of methods respectively and built two kinds of different single sorters;
Step 3-3-1-1: given k-near neighbor method parameter k value, training data sample set (X, Y), wherein X=[x
1, x
2... x
n] be conditional attribute, Y=[y
1, y
2... y
n] be decision attribute;
Step 3-3-1-2: set t as the conditional attribute number of training data sample, for given set of data samples X ' to be assessed=[x '
1, x '
2... x '
m], calculate respectively each data sample to be assessed and all Sensory Quality of Cigarette assess the Euclidean distance between training data sample, wherein Euclidean distance is defined as
Wherein, x
iprepresent training data sample x
ithe value of conditional attribute p; X '
jprepresent data sample x ' to be assessed
jthe value of conditional attribute p;
Step 3-3-1-3: assess the Euclidean distance between training data sample, using the final classification of the mode of the classification of the k nearest apart from this data sample to be assessed training data sample as this data sample to be assessed based on each data sample to be assessed calculated and whole Sensory Quality of Cigarette.
Step 3-3-2: based on decision tree C4.5 method establishment tobacco product sensory evaluating smoking index prediction model, namely based on the single sorter of decision tree C4.5 method;
Step 3-3-2-1: given training data sample set, given decision tree C4.5 method parameter serious forgiveness inc;
The parameter serious forgiveness of decision tree C4.5 represents that each node carries out the end condition of branch's achievement, and to prevent the phenomenon of over-fitting in achievement process, choosing C4.5 serious forgiveness is here 5%;
Step 3-3-2-2: (classification number is m) to the data set setting D to comprise as some nodes, p
ifor the ratio of classification in data set D shared by the sample of i, then define priorentropy Info (D)
If A={a
1, a
2... a
nit is the value set of conditional attribute A; D={D
1, D
2... D
nfor the value by attribute A is to the data set obtained after data set D division, wherein D
jrepresent that in data set D, attribute A value is a
jsubset set, then the conditional entropy Info of definite condition attribute A
a(D)
Wherein p
jifor data set D
jratio shared by middle classification i, | D
j| with | D| represents set D respectively
jwith the total sample number in D, the information gain Gain (A) of definite condition attribute A
Gain(A)=Info(D)-Info
A(D)
Step 3-3-2-3: the whole values traveling through each attribute, and calculate respectively using this value as breakpoint by discrete for attribute A be the discrete data of two classification time information gain value, information gain is worth maximum attribute value as breakpoint, discretize is carried out to each attribute;
Step 3-3-2-4: based on information gain Gain (A), definition information ratio of profit increase GainRatio (A)
Top-downly at the attribute that each node selection makes information gain-ratio maximum, the value after discrete according to it carries out the branch of decision tree;
Step 3-3-2-5: judge whether at each node to meet the end condition of contributing, if do not meet, return step 3-3-2-4 and proceed branch, if meet, using now node comprise the classification of mode as this leaf node of decision attribute in data, then the branch of next node is carried out, until all node all meets branch condition, end condition is defined as
(L/M<inc)∪(U=1)
Wherein L is the data sample number that this node comprises, and M is data sample sum, and U comprises by this node the classification number of data;
Step 3-3-2-6: for given data to be assessed, exports the classification that can obtain leaf node according to decision tree top-downly.
Step 3-3-3: based on support vector machine (SVM) method establishment tobacco product sensory evaluating smoking index prediction model, namely based on the single sorter of support vector machine method;
Step 3-3-3-1: given training data sample set (x
1, y
1), (x
2, y
2) ..., (x
l, y
l), given SVM method kernel function, penalty parameter c, nuclear parameter g;
The kernel function of SVM chooses radial basis (RBF) function, penalty parameter c (in adjustment feature subspace, the fiducial range of SVM model and the ratio of empiric risk make the generalization ability of support vector machine reach best) chooses c=2, and nuclear parameter g (affecting the complexity that sample data distributes in high-dimensional feature space) chooses g=1;
Step 3-3-3-2: by introducing Lagrange function, SVM algorithm is summed up as constrained quadratic programming (QP) problem:
s.t.y
i[(x
i·ω)+b]-1+ξ
i≥0,i=1,2,...,l
Wherein x ω+b=0 is lineoid, ξ
i> 0 represents the slack variable of punishing classification error sample, the constant of A > 0 for balancing between the punishment level that is used for maintaining sample misclassification and causes and algorithm complex, ω is the weight vectors of input variable, b is the threshold value scalar of lineoid, and SVM is intended to searching one optimum lineoid and lineoid both sides point and lineoid distance are maximized;
Step 3-3-3-3: solving based on optimization problem, is converted into the dual problem of QP problem by this problem:
Wherein α
irepresent Lagrange multiplier, K (x
ix
j) be kernel function, those α
ithe training data sample point of > 0 correspondence is called support vector;
Step 3-3-3-4: for given data sample x to be assessed, based on the support vector obtained, calculate classification results, classification function is defined as
Wherein α
i *represent the Lagrange multiplier of > 0, b
*for the classification thresholds scalar of correspondence.
Step 3-3-4: set up tobacco product sensory evaluating smoking index prediction model based on BP neural net method.
Step 3-3-4-1: neural network initialization, given training data sample set (X, Y) given input layer number M, node in hidden layer P, output layer nodes N, initializes weights value v
ijand ω
jk, initialization hidden layer threshold alpha
j, output layer threshold value beta
k, given learning rate t, given neuron activation functions, given iterations S;
BP neural network adopts single hidden layer configuration, and choosing iterations S is 100, and the number of hidden nodes P is 10, and given learning rate t is 0.001, and initial weight value and threshold value are the random number between 0 ~ 1; Activation function is Sigmoid function.
Step 3-3-4-2: calculate hidden layer by weights and activation function and export H and output layer output O;
Step 3-3-4-3: calculate output layer and export O and the actual error exported between Y of training data sample, definition error E is
e
k=y
k-o
kk=1,2,...,N
Step 3-3-4-4: upgrade weighted value v according to network error
ijand ω
jk
ω
jk=ω
jk+th
je
kj=1,2,...,P;k=1,2,...,N
Step 3-3-4-5: upgrade threshold alpha according to error vector E and weighted value
jand β
k
β
k=β
k+e
kk=1,2,...,N
Step 3-3-4-6: judge whether to reach iterations, reach, complete training process, directly substitutes into step 3.4.2 for data to be assessed and calculates output category result; If do not reach, return step 3.4.2 and continue training, until iterations reaches S.
Step 3-4: utilize each single sorter to classify to Sensory Quality of Cigarette assessment training data sample set, using the weight of the classification results accuracy rate of each single sorter as each single sorter, build the Sensory Quality of Cigarette assessment models based on Multi-classifers integrated.
Build the Sensory Quality of Cigarette assessment models based on Multi-classifers integrated based on above-mentioned single sorter, structure as shown in Figure 3.
Step 4: obtain Sensory Quality of Cigarette assessment result.
Specifically: the tobacco product sense organ data to be assessed for current collection calculate sensory evaluating smoking's index classification result of each single sorter respectively, then the weight of single sorter corresponding for each classification is sued for peace, using the classification of classification maximum for weight sum as sensory evaluating smoking's index, i.e. Sensory Quality of Cigarette assessment result.
In order to verify the sensory evaluation's method performance based on Multi-classifers integrated of the present invention, 684 groups of data samples enterprise provided adopt the mode of ten ten folding cross validations to carry out classification experiments after pre-service: total data sample is divided into ten parts by experiment at random, nine parts that get wherein common as training data sample set at every turn, remaining portion is as set of data samples to be assessed, adopt sorting technique calculate classification results and compare with its actual result, obtain classification accuracy rate.Repeat ten such experiments, and the result of average ten times obtains final classification prediction accuracy.The classification prediction accuracy of ten ten folding cross-validation experiments of single sorter and integrated system experimental technique is as shown in table 3.Wherein, MCS represents method of the present invention.
Table 3 sensory evaluation classifies prediction accuracy
Can see from table, in the six kinds of methods used, the inventive method, relative to other Lung biopsy, all has better effect for six cigarette sensory evaluating smoking indexs.The classifying quality of five kinds of sorting techniques and the inventive method more as shown in Figure 3.
Can find out compared with single basic classification device from the accuracy result of classification prediction, multi-classifier integrating method based on training performance weighting has certain advantage on precision of prediction, the effective innovative approach of one of the intelligent sensory evaluation of cigarette can be carried out as tobacco enterprise, support for the product design process of cigarette and sensory evaluation's process provide certain and help.
Claims (3)
1., based on a Sensory Quality of Cigarette appraisal procedure for Multi-classifers integrated, it is characterized in that, comprise the steps:
Step 1: gather tobacco product sense organ data to be assessed, i.e. cigarette chemical composition index;
Cigarette chemical composition index comprises: total reducing sugar amount, reducing sugar, nicotine amount, total volatile alkaline, nitrogen pool, nicotine nitrogen, protein, schmuck value, nitrogen base ratio, chlorinty, potassium content, sugared alkali ratio, ammonia state alkali;
Step 2: the tobacco product sense organ data to be assessed gathered are normalized based on tobacco product sensory evaluation historical data;
Step 3: utilize the Sensory Quality of Cigarette assessment models based on Multi-classifers integrated to carry out Sensory Quality of Cigarette assessment;
The described Sensory Quality of Cigarette assessment models based on Multi-classifers integrated is: according to the historical data of tobacco product sensory evaluation, utilize the different sorting techniques in Data Mining to build the single sorter describing relation between cigarette chemical composition index and sensory evaluating smoking's index classification result respectively, and utilize the Multi-classifers integrated model that described single sorter builds;
Described sensory evaluating smoking's index comprises: gloss, fragrance, harmony, assorted gas, stimulation, pleasant impression;
Step 4: obtain Sensory Quality of Cigarette assessment result.
2. the Sensory Quality of Cigarette appraisal procedure based on Multi-classifers integrated according to claim 1, is characterized in that, the Sensory Quality of Cigarette assessment models based on Multi-classifers integrated described in step 3 is set up as follows:
Step 3-1: the historical data gathering tobacco product sensory evaluation, sets up Sensory Quality of Cigarette assessment training data sample set;
Sensory Quality of Cigarette assessment training data sample set comprises the expert analysis mode result of cigarette chemical composition index and sensory evaluating smoking's index;
Step 3-2: pre-service is carried out to Sensory Quality of Cigarette assessment training data sample set: according to the specific features of aesthetic quality's judgment criteria and data, respectively discretize is carried out to the expert analysis mode result of 6 kinds of sensory evaluating smoking's indexs; Cigarette chemical composition index is normalized simultaneously;
Step 3-3: utilize the different sorting techniques in pretreated data and Data Mining, builds the classification mathematical prediction model being used for describing relation between cigarette chemical composition and sensory evaluating smoking's index classification result, i.e. single sorter;
Different sorting techniques in described Data Mining, comprising: decision tree C4.5 method, BP neural net method, k-near neighbor method and support vector machine method, and wherein k-near neighbor method gets two different k values, obtain two different single sorters;
Step 3-4: utilize each single sorter to classify to Sensory Quality of Cigarette assessment training data sample set, using the weight of the classification results accuracy rate of each single sorter as each single sorter, build the Sensory Quality of Cigarette assessment models based on Multi-classifers integrated.
3. the Sensory Quality of Cigarette appraisal procedure based on Multi-classifers integrated according to claim 1, it is characterized in that, Sensory Quality of Cigarette assessment result is obtained described in step 4, specifically: the tobacco product sense organ data to be assessed for current collection calculate the classification results of each single sorter respectively, then for each classification, calculating output category result in single sorter is such other sorter weight sum, using the classification of classification maximum for weight sum as sensory evaluating smoking's index, i.e. Sensory Quality of Cigarette assessment result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510232615.1A CN104809230A (en) | 2015-05-08 | 2015-05-08 | Cigarette sensory quality evaluation method based on multi-classifier integration |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510232615.1A CN104809230A (en) | 2015-05-08 | 2015-05-08 | Cigarette sensory quality evaluation method based on multi-classifier integration |
Publications (1)
Publication Number | Publication Date |
---|---|
CN104809230A true CN104809230A (en) | 2015-07-29 |
Family
ID=53694052
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510232615.1A Pending CN104809230A (en) | 2015-05-08 | 2015-05-08 | Cigarette sensory quality evaluation method based on multi-classifier integration |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104809230A (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105223303A (en) * | 2015-10-29 | 2016-01-06 | 云南中烟工业有限责任公司 | A kind of method adopting index to characterize cigarette fruital odor characteristic |
CN105242011A (en) * | 2015-10-08 | 2016-01-13 | 中烟施伟策(云南)再造烟叶有限公司 | Method for evaluating inner sense quality of tobacco dusts |
CN105651941A (en) * | 2015-12-31 | 2016-06-08 | 云南中烟工业有限责任公司 | Intelligent cigarette sensory evaluation system based on decomposition-aggregation strategy |
CN107038254A (en) * | 2017-05-04 | 2017-08-11 | 顾杏春 | Cigarette quality monitoring method and device |
CN109858714A (en) * | 2019-04-03 | 2019-06-07 | 杭州安脉盛智能技术有限公司 | Based on pipe tobacco quality inspection index prediction technique, the apparatus and system for improving neural network |
CN110013048A (en) * | 2019-03-15 | 2019-07-16 | 山东中烟工业有限责任公司 | A kind of tobacco structure suitable for ramuscule cigarette determines method and apparatus |
CN110210680A (en) * | 2019-06-11 | 2019-09-06 | 北京农业信息技术研究中心 | A kind of fish body Noninvasive Measuring Method of Freshness and device based on temperature change |
CN110907591A (en) * | 2019-12-13 | 2020-03-24 | 云南中烟工业有限责任公司 | Method for evaluating sensory quality of heated cigarette |
CN111642782A (en) * | 2020-06-05 | 2020-09-11 | 江苏中烟工业有限责任公司 | Tobacco leaf raw material efficacy positioning method based on cigarette formula requirements |
CN111709627A (en) * | 2020-06-05 | 2020-09-25 | 江苏中烟工业有限责任公司 | Method for judging matching degree of efficacy and style characteristics of cigarette brand formula |
CN115543973A (en) * | 2022-09-19 | 2022-12-30 | 北京三维天地科技股份有限公司 | Data quality rule recommendation method based on knowledge spectrogram and machine learning |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1975705A (en) * | 2005-11-28 | 2007-06-06 | 颐中烟草(集团)有限公司 | Cigarette internal quality index extimating method based on regression function estimating SVM |
CN101894269A (en) * | 2010-07-16 | 2010-11-24 | 西安电子科技大学 | Multi-classifier system-based synthetic aperture radar automatic target recognition method |
CN104484678A (en) * | 2015-01-07 | 2015-04-01 | 青岛理工大学 | Method for diagnosing fusion faults of multiple classifiers on basis of fault type classification capacity evaluation matrix |
-
2015
- 2015-05-08 CN CN201510232615.1A patent/CN104809230A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1975705A (en) * | 2005-11-28 | 2007-06-06 | 颐中烟草(集团)有限公司 | Cigarette internal quality index extimating method based on regression function estimating SVM |
CN101894269A (en) * | 2010-07-16 | 2010-11-24 | 西安电子科技大学 | Multi-classifier system-based synthetic aperture radar automatic target recognition method |
CN104484678A (en) * | 2015-01-07 | 2015-04-01 | 青岛理工大学 | Method for diagnosing fusion faults of multiple classifiers on basis of fault type classification capacity evaluation matrix |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105242011B (en) * | 2015-10-08 | 2017-10-03 | 中烟施伟策(云南)再造烟叶有限公司 | In the evaluation method of aesthetic quality in a kind of smashed tobacco powder |
CN105242011A (en) * | 2015-10-08 | 2016-01-13 | 中烟施伟策(云南)再造烟叶有限公司 | Method for evaluating inner sense quality of tobacco dusts |
CN105223303A (en) * | 2015-10-29 | 2016-01-06 | 云南中烟工业有限责任公司 | A kind of method adopting index to characterize cigarette fruital odor characteristic |
CN105651941A (en) * | 2015-12-31 | 2016-06-08 | 云南中烟工业有限责任公司 | Intelligent cigarette sensory evaluation system based on decomposition-aggregation strategy |
CN107038254B (en) * | 2017-05-04 | 2020-08-11 | 顾杏春 | Cigarette quality monitoring method and device |
CN107038254A (en) * | 2017-05-04 | 2017-08-11 | 顾杏春 | Cigarette quality monitoring method and device |
CN110013048A (en) * | 2019-03-15 | 2019-07-16 | 山东中烟工业有限责任公司 | A kind of tobacco structure suitable for ramuscule cigarette determines method and apparatus |
CN109858714A (en) * | 2019-04-03 | 2019-06-07 | 杭州安脉盛智能技术有限公司 | Based on pipe tobacco quality inspection index prediction technique, the apparatus and system for improving neural network |
CN109858714B (en) * | 2019-04-03 | 2021-05-04 | 杭州安脉盛智能技术有限公司 | Tobacco shred quality inspection index prediction method, device and system based on improved neural network |
CN110210680A (en) * | 2019-06-11 | 2019-09-06 | 北京农业信息技术研究中心 | A kind of fish body Noninvasive Measuring Method of Freshness and device based on temperature change |
CN110907591A (en) * | 2019-12-13 | 2020-03-24 | 云南中烟工业有限责任公司 | Method for evaluating sensory quality of heated cigarette |
CN111642782A (en) * | 2020-06-05 | 2020-09-11 | 江苏中烟工业有限责任公司 | Tobacco leaf raw material efficacy positioning method based on cigarette formula requirements |
CN111709627A (en) * | 2020-06-05 | 2020-09-25 | 江苏中烟工业有限责任公司 | Method for judging matching degree of efficacy and style characteristics of cigarette brand formula |
CN115543973A (en) * | 2022-09-19 | 2022-12-30 | 北京三维天地科技股份有限公司 | Data quality rule recommendation method based on knowledge spectrogram and machine learning |
CN115543973B (en) * | 2022-09-19 | 2023-06-13 | 北京三维天地科技股份有限公司 | Data quality rule recommendation method based on knowledge spectrogram and machine learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104809230A (en) | Cigarette sensory quality evaluation method based on multi-classifier integration | |
CN106779087B (en) | A kind of general-purpose machinery learning data analysis platform | |
CN101587546B (en) | Identification method of lower limb action pattern based on supporting vector multivariate classification | |
CN105589806A (en) | SMOTE+Boosting algorithm based software defect tendency prediction method | |
CN104155574A (en) | Power distribution network fault classification method based on adaptive neuro-fuzzy inference system | |
CN110619084B (en) | Method for recommending books according to borrowing behaviors of library readers | |
CN101833671A (en) | Support vector machine-based surface electromyogram signal multi-class pattern recognition method | |
CN106656357B (en) | Power frequency communication channel state evaluation system and method | |
CN104702465A (en) | Parallel network flow classification method | |
CN109165672A (en) | A kind of Ensemble classifier method based on incremental learning | |
CN105046323B (en) | Regularization-based RBF network multi-label classification method | |
CN112580588A (en) | Intelligent flutter signal identification method based on empirical mode decomposition | |
CN104050547A (en) | Non-linear optimization decision-making method of planning schemes for oilfield development | |
CN109726802A (en) | A kind of railway and wind power plant environment wind speed machine learning prediction technique | |
Wozniak et al. | Designing combining classifier with trained fuser—Analytical and experimental evaluation | |
CN108647772A (en) | A method of it is rejected for slope monitoring data error | |
Osowski et al. | Local dynamic integration of ensemble in prediction of time series | |
CN105651941B (en) | A kind of cigarette sense organ intelligent evaluation system based on decomposition aggregation strategy | |
CN101702172A (en) | Data discretization method based on category-attribute relation dependency | |
Chen et al. | Ensemble methods for solar power forecasting | |
Gu et al. | Fuzzy time series forecasting based on information granule and neural network | |
CN103020864B (en) | Corn fine breed breeding method | |
Zhao et al. | Artificial neural networks based on fractal growth | |
Ahmed et al. | Improving prediction of plant disease using k-efficient clustering and classification algorithms | |
Rosalina et al. | Maximal Overlap Discrete Wavelet Transform, Graph Theory And Backpropagation Neural Network In Stock Market Forecasting |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
EXSB | Decision made by sipo to initiate substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20150729 |