CN1769867A - Stoichiometric identification method for plastic type - Google Patents
Stoichiometric identification method for plastic type Download PDFInfo
- Publication number
- CN1769867A CN1769867A CN 200510044724 CN200510044724A CN1769867A CN 1769867 A CN1769867 A CN 1769867A CN 200510044724 CN200510044724 CN 200510044724 CN 200510044724 A CN200510044724 A CN 200510044724A CN 1769867 A CN1769867 A CN 1769867A
- Authority
- CN
- China
- Prior art keywords
- network
- model
- plastic
- plastics
- artificial neural
- 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.)
- Granted
Links
Images
Landscapes
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
Disclosed is a chemical metrology discrimination method for plastic type which comprises: employing optical spectrum information of plastic infrared spectrum characteristic area, selecting characteristic area infrared spectrum as input variable of competitive back propagation artificial nerve network with nonlinear iterative least square variable selecting method; the competitive back propagation artificial nerve network comprises front end back propagation nerve network and back end competitive function layer, taking unit matrix as network training target, wherein the unit matrix having the amount of known plastic resin sample as rank, having training on network connection weight value with known type plastic sample input matrix and analog noise input matrix and establishing network connection; input variable being converted by back propagation network layer and competitive function layer and the most significant neuron being shown as result, therefore plastic type discrimination system being established. The inventive method has the advantages of being rapid and accurate.
Description
Technical field
The present invention relates to a kind of discrimination method to the plastic resin model---the Chemical Measurement discrimination method of plastics model.
Background technology
Plastics are widely used in the industrial and agricultural production every field, also are the foreign trade bulk supply tariffs.Its material property difference of the plastic resin of variety classes, model is bigger, and it is also bigger to be worth difference, and application is different, and therefore, the discriminating of its model is differentiated significant and practical value for the goods of production and processing, foreign trade.
Prior art is to various plastic resins, discriminating as tygon (PE), polypropylene (PP), polystyrene (PS) etc., be to carry out qualitative identification by characteristic peak and bands of a spectrum to the sample infrared spectrum, promptly utilize method of infrared spectrophotometry for the plastics discriminating of classifying, this method is very effective, and simple, convenient and rapid.But with method of infrared spectrophotometry the model in a certain big class plastic resin a lot of difficulties have been differentiated, because the molecular structure difference of different model plastics is very little in the same class plastics, so can not finish this work basically with infra-red sepectrometry.Although existing other modern advanced analysis means come sample is analyzed owing to differentiated that plastic resin character is close, often can not get good identification result, so how in data the representative information of high efficiency extraction be the key of dealing with problems.
Artificial neural network (ANN) is the forward position and the focus of Chemical Measurement research in recent years, obtain remarkable achievement at aspects such as classification, pattern-recognition, prediction of result, especially be difficult to aspect effective nonlinear problem processing that solves distinctive feature is arranged in processing conventional linear technology, be considered to solve the prefered method of nonlinear problems such as classification, pattern-recognition.
The present invention classifies in conjunction with nonlinear iterative partial least square Variables Selection technology (NIPALS) artificial neural network to plastic resin and model is differentiated, has obtained good result.
Summary of the invention
The Chemical Measurement discrimination method that the purpose of this invention is to provide a kind of plastic resin model can only be to the plastic resin discriminating of classifying to remedy prior art, but is difficult to shortcoming that the model in a certain big class plastic resin is differentiated.
Principle of the present invention is to utilize characteristic area (1350～650nm in different the plastic sample infrared spectrums, particularly infrared spectrum
-1) spectral information, adopt the variable input of one section representative sample fingerprint region infrared spectrum of the nonlinear iterative partial least square Variables Selection choice of technology as competitive back-propagation artificial neural network; This competitiveness reverse transmittance nerve network is made of front end reverse transmittance nerve network layer and rear end competitive function layer, with the number of known plastic resin sample is that the unit matrix of order is the network training target, with the plastic sample input matrix of known models and simulate noisy vocal input matrix and network is connected weights train, set up network and connect; Input variable is after the conversion of counterpropagation network layer and competitive function layer, and the most significant neuron is won and as a result of showed, and the plastics model of checking the training objective correspondence obtains final identification result, thereby forms plastics model recognition system.
Discrimination method of the present invention is at first to test the plastic resin sample with Fourier transformation infrared spectrometer, obtains corresponding infrared spectrogram, with 2nm
-1For the ir data in interval unit intercepting diffuse reflectance infrared spectroscopy district, be the sample point space with various plastics models, the ir data of intercepting is that the variable space constitutes the raw data matrix that the plastics model is differentiated; Then with raw data matrix through border standardization processing, obtain the standardized data matrix; By nonlinear iterative partial least square Variables Selection technology the standardized data matrix is carried out principal component analysis (PCA), the major component number adopts ratio method to do discriminant criterion, the major component number of determining the nonlinear iterative partial least square Variables Selection is 3, as the variable input K of artificial neural network.Further the network connection of competitive back-propagation artificial neural network is trained, with the number of wherein plastic sample is that the unit matrix of order is as the network training target, set up network and connect, promptly produce " 1 ", all the other position outputs " 0 " in the output vector relevant position; Output layer is n, network structure is three layers of error back propagation artificial neural network of input layer, hidden layer and output layer, hidden layer contains 10 neurons, (IW (1 with logsig, i) p (i)+b1) function is as transport function, adopt adaptive learning speed to train in conjunction with the momentum mode, the momentum term constant is 0.95, with the total error quadratic sum as convergence criterion; Then counterpropagation network is exported after the conversion of competition layer function, the neuron of triumph as a result of shows, and obtains the preliminary classification result to the input model, forms the elementary recognition system of plastics model.
In order further to improve the accuracy rate that the plastics model is differentiated, the data set that adopts many groups to contain noise again connects network trains again, should organize the data set that contains noise is on the raw data basis more, noise level according to determination data, utilize noise to produce function and produce many group normal distribution noises at random, be formed by stacking, form practical competitive back-propagation artificial neural network after training is finished, thereby set up the recognition system of corresponding artificial neural network identification plastic resin model.
The wave-number range of characteristic area is 1350～650nm in the above-mentioned infrared spectrum
-1
Advantage of the present invention is to differentiate the model of common plastic resin accurately, fast and effectively.
Description of drawings:
Fig. 1 global procedures synoptic diagram of the present invention.
Fig. 2 network training final goal is that the number with known polyethylene specimen is the unit matrix synoptic diagram of order.
Fig. 3 differentiates the used back-propagation artificial neural network layer of plastics model structural representation.
Fig. 4 discrimination natwork training process synoptic diagram
The identification result tabulation figure of vinyon PE203, the PE722 of Fig. 5 embodiment.
The vinyon PE203 infrared spectrogram of Fig. 6 embodiment.
The vinyon PE722 infrared spectrogram of Fig. 7 embodiment.
Wherein, and W (1, i): the corresponding weight value that each input p (i) of hidden layer is corresponding;
W (2, i): the corresponding weight value of each input a1 correspondence of output layer;
P (i): input;
A1: hidden layer output, output layer input;
A2: output layer output;
B1: threshold value;
B2: threshold value;
Logsig (IW (1, i) p (i)+b1): transport function.
N: plastics model number
Embodiment
Present embodiment is that two kinds of plastic resins of PE203, PE722 are example explanation the present invention with the polyvinyl resin model.The present invention is with MatLab
Programming with Pascal Language realizes, at the MatLab of embedded neural network kit
Environment is operation down, and global procedures of the present invention as shown in Figure 1.Concrete steps are as follows:
The first step with the known plastic sample of Fourier transformation infrared spectrometer test, as 300 kinds of polyvinyl resin samples, obtains corresponding infrared spectrogram (infrared spectrogram of PE203, PE722 is seen Fig. 6,7), with 2nm
-1Be the intercepting characteristic area (1350～650nm of interval unit
-1) ir data, be the sample point space with 300 kinds of plastics models, the ir data of intercepting is that the variable space constitutes the raw data matrix that the plastics model is differentiated.
Second step, raw data matrix is carried out standardization, obtain the standardized data matrix, promptly the centralization processing of data and nondimensionalization are handled, and the average of each variable is 0 in its variable space of matrix of standardization, and variance is 1.
The 3rd step, according to nonlinear iterative partial least square Variables Selection technology the standardized data matrix is carried out principal component analysis (PCA), the major component number adopts ratio method to do discriminant criterion, the major component number of determining the nonlinear iterative partial least square Variables Selection is 3, as the variable input K of artificial neural network.
The 4th step connected the network of back-propagation artificial neural network and to train, and was that the unit matrix of order is the network training target with the number of known 300 kinds of vinyon samples, setting up network connects, as Fig. 2, promptly produce " 1 " all the other position outputs " 0 " in the output vector relevant position; Output layer is n (n is a polyethylene standard sample number, and n is 300 in the present embodiment) output, sees Fig. 2.Network structure is three layers of (input layer, hidden layer and output layer) back-propagation artificial neural network, hidden layer contains 10 neurons, (IW (1 to adopt logsig, i) p (i)+b1) function is as transport function, adopt adaptive learning speed to train in conjunction with the momentum mode, the momentum term constant is 0.95, with the total error quadratic sum as convergence criterion.Counterpropagation network is exported after the conversion of competition layer function, the output of the easiest generation " 1 " is by competition layer competition-inhibiting effect, only an input becomes the victor at last, other is output as 0, to relevant each connection weight of triumph neuron towards the direction adjustment that more helps competing, the neuron of winning as a result of shows, obtain the noiseless matrix, this process is the raw data training process, obtain preliminary classification result like this, form the elementary recognition system of plastics model the input model.The back-propagation artificial neural network structural drawing that plastic sample is differentiated is seen Fig. 3.
The 5th step, in order further to improve the accuracy rate that the plastics model is differentiated, by on the basis of the raw data training that obtains of step, the data set that adopts many groups to contain noise again connects network trains.The data set that contains noise is on the basis of raw data set, noise level according to determination data, utilize noise to produce function and produce many group normal distribution noises at random, formation superposes, form practical competitive back-propagation artificial neural network, thereby set up the recognition system of the effective using artificial neural networks identification of cover plastic resin model.Contain noise discrimination natwork training process and see synoptic diagram 4.
The 6th step, when the plastic sample of the unknown is differentiated, input is through the plastics of the known models of Variables Selection such as above-mentioned tygon masterbatch PE203, the characteristic area spectroscopic data of PE722, after the conversion of counterpropagation network and competitive function layer, the most significant neuron just can be won and as a result of be showed, be boolean's output that corresponding vector element position produces " 1 ", other position generation " 0 ", as shown in Figure 5, check the plastics model of training objective correspondence, obtain plastics model identification result.Form practical competitive back-propagation artificial neural network after training is finished, thereby set up the recognition system of corresponding artificial neural network identification plastic resin model.
Utilize above-mentioned method or step, can distinguish the recognition system of corresponding practical artificial neural network identification plastic resin model, and 18 common big classes, more than 1000 kind of plastic resin model have been set up corresponding recognition system.
18 common big class plastic resins comprise: polyvinyl resin PE, acrylic resin PP, polystyrene resin PS, acrylonitrile-styrene resin AS, acrylonitrile-styrene-acrylic ester Resin A SA, vinyl-vinyl acetate copolymer EVA, polyethylene terephthalate PET, polybutylene terephthalate PBT, polymethyl methacrylate resin PMMA, polyformaldehyde resin POM, polyolefin plastomers POPs, polyphenylene oxide resin PPO, polyurethane resin PU, Corvic PVC, liquid crystal polymer LCP, polyamide PA, the ionic polymerization resin, other functional plasticses.The present invention is recognition objective and training set with 18 common big classes, more than 1000 kind of plastic resin model, set up the recognition system of corresponding practical artificial neural network identification plastic resin model respectively, utilized this system to differentiate that the plastic resin model fast, accurately, reliably.
Claims (4)
1, a kind of Chemical Measurement discrimination method of plastics model is characterized in that at first testing the plastic resin sample with Fourier transformation infrared spectrometer, obtains corresponding infrared spectrogram, with 2nm
-1For the ir data in interval unit intercepting diffuse reflectance infrared spectroscopy district, be the sample point space with various plastics models, the ir data of intercepting is that the variable space constitutes the raw data matrix that the plastics model is differentiated; With the standardization of raw data matrix process, obtain the standardized data matrix then; By nonlinear iterative partial least square Variables Selection technology the standardized data matrix is carried out principal component analysis (PCA), the major component number adopts ratio method to do discriminant criterion, the major component number of determining the nonlinear iterative partial least square Variables Selection is 3, variable input as artificial neural network, further the network connection of competitive back-propagation artificial neural network is trained, with the number of wherein plastic sample is that the unit matrix of order is as the network training target, setting up network connects, promptly produce " 1 " all the other position outputs " 0 " in the output vector relevant position; Output layer is n, network structure is three layers of error back propagation artificial neural network of input layer, hidden layer and output layer, hidden layer contains 10 neurons, (IW (1 with logsig, i) p (i)+b1) function is as transport function, adopt adaptive learning speed to train in conjunction with the momentum mode, the momentum term constant is 0.95, with the total error quadratic sum as convergence criterion; Then counterpropagation network is exported after the conversion of competition layer function, the neuron of triumph as a result of shows, and obtains the preliminary classification result of input model in pairs, the elementary recognition system of formation plastics model.
2, the Chemical Measurement discrimination method of plastics model as claimed in claim 1, the wave-number range that it is characterized in that characteristic area in the above-mentioned infrared spectrum is 1350～650nm
-1
3, the Chemical Measurement discrimination method of plastics model as claimed in claim 1, it is characterized in that on the elementary recognition system basis of plastics model, adopt many groups to contain the data set of noise again to network connection carrying out retraining, should organize the data set that contains noise is on the raw data basis more, noise level according to determination data, utilize noise to produce function and produce many group normal distribution noises at random, be formed by stacking, form practical competitive back-propagation artificial neural network after training is finished, thereby set up the recognition system of corresponding artificial neural network identification plastic resin model.
4, the Chemical Measurement discrimination method of plastics model as claimed in claim 3 is characterized in that above-mentioned noise produces function and produces at random how the group noises are normal distribution noises.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2005100447247A CN100472200C (en) | 2005-09-16 | 2005-09-16 | Stoichiometric identification method for plastic type |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2005100447247A CN100472200C (en) | 2005-09-16 | 2005-09-16 | Stoichiometric identification method for plastic type |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1769867A true CN1769867A (en) | 2006-05-10 |
| CN100472200C CN100472200C (en) | 2009-03-25 |
Family
ID=36751292
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB2005100447247A Expired - Fee Related CN100472200C (en) | 2005-09-16 | 2005-09-16 | Stoichiometric identification method for plastic type |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN100472200C (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101221165B (en) * | 2008-01-24 | 2011-05-04 | 大连工业大学 | Portable plastic and polymer material main composition detecting instrument |
| CN102175637A (en) * | 2010-12-30 | 2011-09-07 | 中国药品生物制品检定所 | Method for detecting plastics |
| CN105372287A (en) * | 2015-12-22 | 2016-03-02 | 天津市建筑材料产品质量监督检测中心 | Method for detecting polystyrene reworked material in extruded polystyrene board |
| CN105574587A (en) * | 2016-01-21 | 2016-05-11 | 华中科技大学 | On-line condition process monitoring method for plastic injection moulding process |
| CN114965973A (en) * | 2022-05-12 | 2022-08-30 | 知里科技(广东)有限公司 | Method for identifying recycled plastic based on instrument detection and analysis technology combined with multiple chemometrics methods and/or machine learning algorithm |
-
2005
- 2005-09-16 CN CNB2005100447247A patent/CN100472200C/en not_active Expired - Fee Related
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101221165B (en) * | 2008-01-24 | 2011-05-04 | 大连工业大学 | Portable plastic and polymer material main composition detecting instrument |
| CN102175637A (en) * | 2010-12-30 | 2011-09-07 | 中国药品生物制品检定所 | Method for detecting plastics |
| CN102175637B (en) * | 2010-12-30 | 2012-12-19 | 中国药品生物制品检定所 | Method for detecting plastics |
| CN105372287A (en) * | 2015-12-22 | 2016-03-02 | 天津市建筑材料产品质量监督检测中心 | Method for detecting polystyrene reworked material in extruded polystyrene board |
| CN105372287B (en) * | 2015-12-22 | 2017-12-19 | 天津市建筑材料产品质量监督检测中心 | The detection method of polystyrene reworked material in a kind of extruded polystyrene board |
| CN105574587A (en) * | 2016-01-21 | 2016-05-11 | 华中科技大学 | On-line condition process monitoring method for plastic injection moulding process |
| CN105574587B (en) * | 2016-01-21 | 2017-03-08 | 华中科技大学 | A kind of online operating mode course monitoring method of plastic injection molding process |
| CN114965973A (en) * | 2022-05-12 | 2022-08-30 | 知里科技(广东)有限公司 | Method for identifying recycled plastic based on instrument detection and analysis technology combined with multiple chemometrics methods and/or machine learning algorithm |
Also Published As
| Publication number | Publication date |
|---|---|
| CN100472200C (en) | 2009-03-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN100472200C (en) | Stoichiometric identification method for plastic type | |
| CN1381721A (en) | Portable intelligent electronic nose and its preparing process | |
| JP4581039B2 (en) | Grade identification method for polymer materials | |
| CN113705715B (en) | Time sequence classification method based on LSTM and multi-scale FCN | |
| US20230243744A1 (en) | Method and system for automatically detecting and reconstructing spectrum peaks in near infrared spectrum analysis of tea | |
| CN116204831A (en) | Road-to-ground analysis method based on neural network | |
| CN114399674A (en) | Hyperspectral image technology-based shellfish toxin nondestructive rapid detection method and system | |
| Dan | NIR spectroscopy oranges origin identification framework based on machine learning | |
| Broek et al. | Application of a spectroscopic infrared focal plane array sensor for on-line identification of plastic waste | |
| CN1198261C (en) | Voice identification based on decision tree | |
| CN116519661A (en) | Rice identification detection method based on convolutional neural network | |
| Jin et al. | Anti-corrosion wood automatic sorting robot system based on near-infrared imaging technology | |
| CN113378935B (en) | Intelligent olfactory sensation identification method for gas | |
| CN111950427B (en) | Garnet subclass identification method based on thermal infrared spectrum characteristics and BP neural network model | |
| CN114863286A (en) | Mixed waste plastic classification method based on multi-algorithm collaborative optimization | |
| CN115049006A (en) | Communication signal identification method and system based on self-adaptive feedforward neural network | |
| CN114742091A (en) | Method, system and medium for identifying radar individual radiation based on convolution block attention | |
| Fajri et al. | The recognition of mango varieties based on the leaves shape and texture using back propagation neural network method | |
| US20040042665A1 (en) | Method and computer program product for automatically establishing a classifiction system architecture | |
| CN103411913B (en) | A kind of fish oil infrared spectrum PLS recognition methods based on the adaptively selected waypoint of genetic algorithm | |
| CN118366035B (en) | Textile fabric hyperspectral component sensing method and device | |
| Sapkal et al. | Analysis of classification by supervised and unsupervised learning | |
| Jia et al. | Gas Composition Recognition Based on Analyzing Acoustic Relaxation Absorption Spectra: Wavelet Decomposition and Support Vector Machine Classifier | |
| CN114492572B (en) | Material structure classification method and system based on machine learning clustering algorithm | |
| CN1282106C (en) | Rapid study classifying method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| C17 | Cessation of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20090325 Termination date: 20100916 |