CN105005790B - Poison gas intelligent identification Method in electronic nose room based on semi-supervised learning - Google Patents
Poison gas intelligent identification Method in electronic nose room based on semi-supervised learning Download PDFInfo
- Publication number
- CN105005790B CN105005790B CN201510391640.4A CN201510391640A CN105005790B CN 105005790 B CN105005790 B CN 105005790B CN 201510391640 A CN201510391640 A CN 201510391640A CN 105005790 B CN105005790 B CN 105005790B
- Authority
- CN
- China
- Prior art keywords
- data set
- classification device
- label
- poison gas
- mrow
- 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.)
- Expired - Fee Related
Links
Classifications
-
- 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
- G06F18/21—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
Abstract
The invention discloses poison gas intelligent identification Method in a kind of electronic nose room based on semi-supervised learning,Using each basic classification device of the poison gas sample data set L sample training of known label,Each basic classification device is alternately as Main classification device in each study circulation,Unknown Label sample data set U is classified by Main classification device,And the label of data in sample data set U is predicted using remaining basic classification device,In result is voted,If the votes of a certain data label exceed the threshold value pre-set in data set U,Then the sample data will be used to and original data set L re -training graders together with its label,Finally again by increasing the number of grader come whether the discrimination of decision-making system has been optimal,Grader after so training not only has more basic classification device scales,And there is the stronger ability from unknown exemplar learning smell pattern.
Description
Technical field
The present invention relates to the Classification and Identification technology in e-nose signal processing, is that one kind is based on semi-supervised specifically
Poison gas intelligent identification Method in the electronic nose room of habit.
Background technology
At present, for indoor detection of poison gas, in order to ensure the correctness of testing result, used electric nasus system is being instructed
Substantial amounts of learning sample need to be used by practicing the stage, and in general, poison gas is divided using the electronic nose for having label data to train to obtain
Class accuracy is higher than based on the electronic nose for training to obtain without label data, but without label data than there is label data more to hold
Easily obtain.
It is thus proposed that semi-supervised learning technology, by semi-supervised learning technology can help electronic nose not only from
Training sample learning associative mode, also can be from unknown exemplar learning relevant knowledge, so as to realize to certain smell mould
The continuous learning of formula is i.e. untill any raising no longer occurs for discrimination.
But, it has been suggested that semi-supervised learning technology existing for deficiency:First, a big chunk semi-supervised learning technology pin
To be two classification problems, and the species of indoor poison gas is far above two classes, therefore can not meet application demand;Second, it can carry out more
The semi-supervised learning algorithm of classification, its grader scale are defined again, cause the results of learning to test sample bad, classification essence
Degree is not high.
The content of the invention
In view of the shortcomings of the prior art, it is an object of the invention to provide malicious in a kind of electronic nose room based on semi-supervised learning
Gas intelligent identification Method, basic classification device scale is the method increase, had stronger from unknown exemplar learning smell
The ability of pattern, pattern-recognition rate of the electronic nose to each poison gas is enabled to reach optimal level.
To reach above-mentioned purpose, the technical solution adopted by the present invention is as follows:
Poison gas intelligent identification Method in a kind of electronic nose room based on semi-supervised learning, its key are according to following steps
Carry out:
Step 1:The poison gas sample data set L of known label and the poison gas sample data set U of Unknown Label are obtained, presets base
The number M=3 of this grader, current frequency of training are t;
Step 2:The equal subset L of M scale is randomly generated from the poison gas sample data set L of known labeliTo train
Each basic classification device ci, i=1~M;
Step 3:The poison gas sample data set L of known label is carried out using each basic classification device that step 2 trains
Classification and Identification, the initial identification rate of each grader is obtained, carried out using differentiation result of the simple vote method to all graders
Integrate, obtain system initial identification rate;
Step 4:If i-th of basic classification device ciFor main grader, the poison gas sample using Main classification device to Unknown Label
Data in data set U are classified, and the poison gas sample data set using remaining M-1 basic classification device to Unknown Label
The label of data is predicted in U, obtains prediction error rate ei(t);
Step 5:As the prediction error rate e for the basic classification device that this is trainedi(t) it is less than last prediction error rate ei
(t-1) when, if the data in the poison gas sample data set U of Unknown Label are by the result of remaining M-1 basic classification device ballot
More than default threshold θ, then the data are incorporated to data set Li(t) in;
Step 6:Judge whether to meetWherein | Li(t) | represent this training dataset
Li(t) scale, | Li(t-1) | represent last training dataset Li(t-1) scale, ei(t) this base trained is represented
This grader ciPrediction error rate, ei(t-1) the last basic classification device c trained is representediPrediction error rate;
If it is satisfied, then utilize the new data set L obtained by step 5iAnd original data subset L (t)iTo basic classification
Device ciCarry out re -training;
Otherwise, from the new data set L obtained by step 5i(t) it is sub with original data again after s sample of random removal in
Collect LiTo basic classification device ciRe -training is carried out, wherein:Int () is to take
Integral function;
Step 7:Operation of the step 4 to step 6 is carried out to M basic classification device successively according to i=1~M, until each base
The discrimination of this grader no longer changes;
Step 8:According to the number of M=M+1 increase basic classification devices, the operation of repeat step 2 to step 7, until system
Discrimination reach target.
In implementation process, the basic classification device is built using SVMs or artificial neural network, also may be used certainly
With using other classification and identification algorithms structure basic classification device.
Preferably, in step 4 according toCalculate prediction error rate, wherein ni(t) represent in t
During secondary training, predicted in the poison gas sample data set U of Unknown Label by remaining M-1 basic classification device and obtain the sample of label
This, ni' (t) represent to be predicted by remaining M-1 basic classification device in the poison gas sample data set U of Unknown Label and obtain correct mark
The sample of label.
In the present invention, using each basic classification device of the poison gas sample data set L sample training of known label, every
Each basic classification device is alternately as Main classification device in secondary study circulation, by Main classification device to Unknown Label sample data set
U is classified, and the label of data in sample data set U is predicted using remaining basic classification device, is being voted
As a result in, if the votes of a certain data label exceed the threshold value pre-set in data set U, the sample data connects
The re -training grader together with original data set L will be used to its label, finally again by increasing the number of grader
Whether the discrimination that mesh carrys out decision-making system has been optimal, and the grader after so training not only has more basic classifications
Device scale, and there is the stronger ability from unknown exemplar learning smell pattern.
The present invention remarkable result be:Compared to existing classification and identification algorithm, basic classification device scale is not only increased, and
And there is the stronger ability from unknown exemplar learning smell pattern so that the electronic nose finally obtained is to each poison
The pattern-recognition rate of gas reaches optimal level.
Embodiment
The embodiment and operation principle of the present invention are described in further detail below.
Poison gas intelligent identification Method in a kind of electronic nose room based on semi-supervised learning, is followed the steps below:
Step 1:The poison gas sample data set L of known label and the poison gas sample data set U of Unknown Label are obtained, presets base
The number M=3 of this grader, current frequency of training are t;
Step 2:The equal subset L of M scale is randomly generated from the poison gas sample data set L of known labeliTo train
Each basic classification device ci, i=1~M;
Step 3:The poison gas sample data set L of known label is carried out using each basic classification device that step 2 trains
Classification and Identification, the initial identification rate of each grader is obtained, carried out using differentiation result of the simple vote method to all graders
Integrate, obtain system initial identification rate;
Step 4:If i-th of basic classification device ciFor main grader, the poison gas sample using Main classification device to Unknown Label
Data in data set U are classified, and the poison gas sample data set using remaining M-1 basic classification device to Unknown Label
The label of data is predicted in U, obtains prediction error rate ei(t);
Step 5:As the prediction error rate e for the basic classification device that this is trainedi(t) it is less than last prediction error rate ei
(t-1) when, if the data in the poison gas sample data set U of Unknown Label are by the result of remaining M-1 basic classification device ballot
More than default threshold θ, then the data are incorporated to data set Li(t) in;
Step 6:Judge whether to meetWherein | Li(t) | represent this training dataset
Li(t) scale, | Li(t-1) | represent last training dataset Li(t-1) scale, ei(t) this base trained is represented
This grader ciPrediction error rate, ei(t-1) the last basic classification device c trained is representediPrediction error rate;
If it is satisfied, then utilize the new data set L obtained by step 5iAnd original data subset L (t)iTo basic classification
Device ciCarry out re -training;
Otherwise, from the new data set L obtained by step 5i(t) it is sub with original data again after s sample of random removal in
Collect LiTo basic classification device ciRe -training is carried out, wherein:Int () is
Bracket function;
Step 7:Operation of the step 4 to step 6 is carried out to M basic classification device successively according to i=1~M, until each base
The discrimination of this grader no longer changes;
Step 8:According to the number of M=M+1 increase basic classification devices, the operation of repeat step 2 to step 7, until system
Discrimination reach target.
In the present embodiment, the basic classification device is built using SVMs.
In step 4 according toCalculate prediction error rate, wherein ni(t) represent when training for the t times,
Predicted in the poison gas sample data set U of Unknown Label by remaining M-1 basic classification device and obtain the sample of label, ni' (t) table
Show and predicted in the poison gas sample data set U of Unknown Label by remaining M-1 basic classification device and obtain the sample of correct label.But
It is because having no idea to calculate the classification accuracy rate of the sample set of Unknown Label, and only data set L is that available (label is
Know, classification accuracy rate can be calculated), thus it is false as same data distribution based on thering are exemplar and unlabeled exemplars to meet
If premise, in specific implementation process, data set U discrimination is replaced using data set L discrimination, it is corresponding to use data set
L prediction error rate goes the prediction error rate instead of data set U.
For a further understanding of the technique effect of the present invention, the final classification device obtained separately below using present invention training
And bibliography:Z-H.Zhou,Tri-training:exploiting unlabeled data using three
classifiers,Knowledge and Data Engineering,IEEE Transactions on,17(2005)1529-
The grader that the 1541. Tri-training Algorithm for Training that are proposed obtain to three class indoor pollution poison gases (benzene, toluene,
Formaldehyde) Classification and Identification is carried out, recognition result is as shown in table 1.
The Tri-training of table 1 and the discrimination for carrying not test set corresponding to this programme of same amount basic classification device
(%)
Remarks:
1st, basic classification device (is only an example, it is not limited to which this, artificial neural network etc. is equal using SVMs
Can);
2nd, discrimination herein is test set discrimination, is totally independent of data set L and U, for verifying that half supervises
Educational inspector practises the validity that poison gas discrimination accuracy is carried out to lifting electronic nose;
3rd, Classification accuracy (initial) are obtained in step 3, will be all using simple vote method
The result of basic classification device is collected;
4th, Classification accuracy (final) are obtained in step 8, calculation with
Classification accuracy (initial) are identical;
5th, Impro=(final accuracy-initial accuracy)/initial accuracy.
It can be seen that in any case from the comparing result of table 1, employ more basic classification devices to mean have more
Chance can be acquired from the sample of Unknown Label, so as to improve recognition correct rate;And from respectively with 4 basic point
The recognition result of the inventive method of class device, 5 basic classification devices and 6 basic classification devices shows, when Unknown Label sample set
After it is determined that, the knowledge that the sample set can provide is exactly limited, then improves the quantity of basic classification device and also can not infinitely improve
Discrimination, now only expand the scale of unknown sample collection, could further improve discrimination.
In summary, the present invention is in e-nose signal identification process, by combining existing semi-supervised learning method, simultaneously
The discrimination of system is improved by constantly extending grader number, for indoor detection of poison gas, is increased to a certain extent
From the chance of unknown exemplar learning knowledge, indoor detection of poison gas can be effectively realized.
Claims (2)
1. poison gas intelligent identification Method in a kind of electronic nose room based on semi-supervised learning, it is characterised in that enter according to following steps
OK:
Step 1:The poison gas sample data set L of known label and the poison gas sample data set U of Unknown Label are obtained, presets basic point
The number M=3 of class device, current frequency of training are t;
Step 2:The equal subset L of M scale is randomly generated from the poison gas sample data set L of known labeliTo train each base
This grader ci, i=1~M;
Step 3:The poison gas sample data set L of known label is classified using each basic classification device that step 2 trains
Identification, is obtained the initial identification rate of each grader, is integrated using differentiation result of the simple vote method to all graders,
Obtain system initial identification rate;
Step 4:If i-th of basic classification device ciFor main grader, the poison gas sample data set using Main classification device to Unknown Label
Data in U are classified, and using remaining M-1 basic classification device to number in the poison gas sample data set U of Unknown Label
According to label be predicted, obtain prediction error rate ei(t):
<mrow>
<msub>
<mi>e</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mrow>
<msub>
<mi>n</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<msup>
<msub>
<mi>n</mi>
<mi>i</mi>
</msub>
<mo>&prime;</mo>
</msup>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msub>
<mi>n</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
</mrow>
Wherein ni(t) represent when training for the t times, by remaining M-1 basic classification in the poison gas sample data set U of Unknown Label
Device is predicted and obtains the sample of label, ni' (t) is represented in the poison gas sample data set U of Unknown Label by basic point of remaining M-1
Class device is predicted and obtains the sample of correct label;
Step 5:As the prediction error rate e for the basic classification device that this is trainedi(t) it is less than last prediction error rate ei(t-1)
When, if the data in the poison gas sample data set U of Unknown Label are exceeded in advance by the result of remaining M-1 basic classification device ballot
If threshold θ, then the data are incorporated to data set Li(t) in;
Step 6:Judge whether to meetWherein | Li(t) | represent this training dataset Li(t)
Scale, | Li(t-1) | represent last training dataset Li(t-1) scale, ei(t) basic point that this is trained is represented
Class device ciPrediction error rate, ei(t-1) the last basic classification device c trained is representediPrediction error rate;
If it is satisfied, then utilize the new data set L obtained by step 5iAnd original data subset L (t)iTo basic classification device ci
Carry out re -training;
Otherwise, from the new data set L obtained by step 5i(t) at random remove s sample after again with original data subset LiIt is right
Basic classification device ciRe -training is carried out, wherein:Int () is to round letter
Number;
Step 7:Operation of the step 4 to step 6 is carried out to M basic classification device successively according to i=1~M, until each basic point
The discrimination of class device no longer changes;
Step 8:According to the number of M=M+1 increase basic classification devices, the operation of repeat step 2 to step 7, until the knowledge of system
Not rate reaches target.
2. poison gas intelligent identification Method in the electronic nose room according to claim 1 based on semi-supervised learning, its feature exist
In:The basic classification device is built using SVMs or artificial neural network.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510391640.4A CN105005790B (en) | 2015-07-06 | 2015-07-06 | Poison gas intelligent identification Method in electronic nose room based on semi-supervised learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510391640.4A CN105005790B (en) | 2015-07-06 | 2015-07-06 | Poison gas intelligent identification Method in electronic nose room based on semi-supervised learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105005790A CN105005790A (en) | 2015-10-28 |
CN105005790B true CN105005790B (en) | 2018-01-09 |
Family
ID=54378455
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510391640.4A Expired - Fee Related CN105005790B (en) | 2015-07-06 | 2015-07-06 | Poison gas intelligent identification Method in electronic nose room based on semi-supervised learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105005790B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108388929A (en) * | 2018-03-27 | 2018-08-10 | 四川大学 | Client segmentation method and device based on cost-sensitive and semisupervised classification |
CN110222171A (en) * | 2019-05-08 | 2019-09-10 | 新华三大数据技术有限公司 | A kind of application of disaggregated model, disaggregated model training method and device |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101135689A (en) * | 2007-09-21 | 2008-03-05 | 华中科技大学 | Electric nose development platform |
CN102866179A (en) * | 2012-09-13 | 2013-01-09 | 重庆大学 | Online recognition and inhibition method based on non-target interference smell in electronic nose of artificial intelligent learning machine |
CN103412003A (en) * | 2013-08-21 | 2013-11-27 | 电子科技大学 | Gas detection method based on self-adaption of semi-supervised domain |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7230528B2 (en) * | 2005-09-20 | 2007-06-12 | Lawrence Kates | Programmed wireless sensor system |
-
2015
- 2015-07-06 CN CN201510391640.4A patent/CN105005790B/en not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101135689A (en) * | 2007-09-21 | 2008-03-05 | 华中科技大学 | Electric nose development platform |
CN102866179A (en) * | 2012-09-13 | 2013-01-09 | 重庆大学 | Online recognition and inhibition method based on non-target interference smell in electronic nose of artificial intelligent learning machine |
CN103412003A (en) * | 2013-08-21 | 2013-11-27 | 电子科技大学 | Gas detection method based on self-adaption of semi-supervised domain |
Non-Patent Citations (1)
Title |
---|
"多点测量式毒气阵列检测系统研究与光谱分析";廖海洋 等;《光电工程》;20090131;第36卷(第1期);第68-72页 * |
Also Published As
Publication number | Publication date |
---|---|
CN105005790A (en) | 2015-10-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN101216998B (en) | An urban traffic flow information amalgamation method of evidence theory based on fuzzy rough sets | |
CN104156734B (en) | A kind of complete autonomous on-line study method based on random fern grader | |
CN103325122B (en) | Based on the pedestrian retrieval method of Bidirectional sort | |
CN109324604A (en) | A kind of intelligent train resultant fault analysis method based on source signal | |
CN106790019A (en) | The encryption method for recognizing flux and device of feature based self study | |
CN103761507B (en) | Local multi-value pattern face recognition method based on Weber law | |
CN107204114A (en) | A kind of recognition methods of vehicle abnormality behavior and device | |
CN105320966A (en) | Vehicle driving state recognition method and apparatus | |
CN109299741A (en) | A kind of network attack kind identification method based on multilayer detection | |
CN101739337B (en) | Method for analyzing characteristic of software vulnerability sequence based on cluster | |
CN102156887A (en) | Human face recognition method based on local feature learning | |
CN102254183B (en) | Face detection method based on AdaBoost algorithm | |
CN107563274A (en) | A kind of vehicle checking method and method of counting of the video based on confrontation e-learning | |
CN109389325B (en) | Method for evaluating state of electronic transformer of transformer substation based on wavelet neural network | |
CN108009574B (en) | Track fastener detection method | |
CN105740914A (en) | Vehicle license plate identification method and system based on neighboring multi-classifier combination | |
CN110288823B (en) | Traffic violation misjudgment identification method based on naive Bayesian network | |
CN101251896B (en) | Object detecting system and method based on multiple classifiers | |
CN103903441A (en) | Road traffic state distinguishing method based on semi-supervised learning | |
CN103310235B (en) | A kind of steganalysis method based on parameter identification and estimation | |
CN103839080A (en) | Video streaming anomalous event detecting method based on measure query entropy | |
CN109344886A (en) | Occlusion number plate distinguishing method based on convolutional neural network | |
CN102360434B (en) | Target classification method of vehicle and pedestrian in intelligent traffic monitoring | |
CN113378990A (en) | Traffic data anomaly detection method based on deep learning | |
CN107895171A (en) | A kind of intrusion detection method based on K averages Yu depth confidence network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180109 Termination date: 20180706 |
|
CF01 | Termination of patent right due to non-payment of annual fee |