CN109635650A - The recognition methods of the nucleic type of gamma-spectrometric data - Google Patents
The recognition methods of the nucleic type of gamma-spectrometric data Download PDFInfo
- Publication number
- CN109635650A CN109635650A CN201811312752.6A CN201811312752A CN109635650A CN 109635650 A CN109635650 A CN 109635650A CN 201811312752 A CN201811312752 A CN 201811312752A CN 109635650 A CN109635650 A CN 109635650A
- Authority
- CN
- China
- Prior art keywords
- gamma
- spectrometric data
- nucleic
- sample
- classification
- 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
- 238000000034 method Methods 0.000 title claims abstract description 48
- 238000001228 spectrum Methods 0.000 claims abstract description 72
- 238000003066 decision tree Methods 0.000 claims abstract description 64
- 239000013598 vector Substances 0.000 claims abstract description 58
- 238000000605 extraction Methods 0.000 claims abstract description 12
- 239000011159 matrix material Substances 0.000 claims description 32
- 238000012549 training Methods 0.000 claims description 18
- 238000012935 Averaging Methods 0.000 claims description 13
- 238000012360 testing method Methods 0.000 claims description 9
- 238000000342 Monte Carlo simulation Methods 0.000 claims description 6
- 239000004744 fabric Substances 0.000 claims description 2
- 238000013480 data collection Methods 0.000 description 7
- 239000000284 extract Substances 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000015572 biosynthetic process Effects 0.000 description 2
- 238000002790 cross-validation Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000003595 spectral effect Effects 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- -1 and enabling t is 10 Substances 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000005251 gamma ray Effects 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000005658 nuclear physics Effects 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000031068 symbiosis, encompassing mutualism through parasitism Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
Abstract
The invention discloses a kind of recognition methods of the nucleic type of gamma-spectrometric data, comprising: pre-processes to gamma-spectrometric data, to obtain the gamma-spectrometric data being just distributed very much；Characteristics of energy spectrum vector characteristics are extracted in the gamma-spectrometric data being just distributed very much；Described eigenvector feature is established into decision tree classifier as the input vector feature of disaggregated model and carries out classification prediction；The classification predictablity rate is promoted using AdaBoost algorithm.By extracting characteristics of energy spectrum vector characteristics, and the characteristics of energy spectrum vector characteristics according to extraction establish decision tree classifier and carry out classification prediction, then promote the classification predictablity rate using AdaBoost algorithm.The False Rate that single-wheel decision tree classification result is promoted using AdaBoost algorithm, to greatly improve the classification accuracy to nucleic type.
Description
Technical field
The present invention relates to nuclear industry fields, and in particular, to a kind of recognition methods of the nucleic type of gamma-spectrometric data.
Background technique
Currently, parsing gamma-spectrometric data, is the research hotspot that artificial intelligence is applied in industrial circle to nuclide classification identification,
Nuclear industry field, nuclide identification tool have significant practical applications.In a variety of measurement environment, radionuclide passes through gamma energy spectrometer
Etc. nuclear physics equipment and instrument identified, detector receive transmitting gamma ray projector generate gamma-spectrometric data, power spectrum obtain after pass through
Specific method is analyzed, nuclear detection Spectra Unfolding Methods traditional at present, is generallyd use full energy peak method, gradually poor is drawn the technologies such as method
To carry out Radionuclide analysis spectrum unscrambling.
Since the spectrum unscrambling of traditional gamma spectrum information has part special in the presence of that cannot cover whole peak values, the information of power spectrum
Sign, or there is overlapping peak value, the nucleic type so as to cause power spectrum to parse occurs missing, judges result by accident.In addition, right
Change in power spectrum with the dynamic of measurement environmental change, conventional method does not have strong adaptability, and nuclide identification rate can reduce.
Summary of the invention
It is an object of the present invention in view of the above-mentioned problems, propose a kind of recognition methods of the nucleic type of gamma-spectrometric data, with
Realize at least part of solution problems of the prior art.
To achieve the above object, technical solution of the present invention provides
A kind of recognition methods of the nucleic type of gamma-spectrometric data, comprising:
Original gamma-spectrometric data is pre-processed, to obtain the gamma-spectrometric data matrix being just distributed very much；
Dimension-reduction treatment is carried out to the gamma-spectrometric data matrix being just distributed very much, to extract characteristics of energy spectrum vector characteristics；
Decision tree classifier is established to power spectrum number using described eigenvector feature as the input vector feature of disaggregated model
According to nucleic carry out classification prediction；
The classification predictablity rate of the nucleic of the gamma-spectrometric data is promoted using AdaBoost algorithm.
Preferably, the gamma-spectrometric data being just distributed very much, specifically:
Standard deviation is 1, the gamma-spectrometric data that the standard that mean value is zero is just being distributed very much.
Preferably, described that original gamma-spectrometric data is pre-processed, to obtain the gamma-spectrometric data being just distributed very much, comprising:
Generate original gamma-spectrometric data sample；
To the original gamma-spectrometric data sample moment array
The gamma-spectrometric data sample after matrixing is standardized；
Zero averaging is carried out to the gamma-spectrometric data sample after standardization, to obtain the gamma-spectrometric data matrix being just distributed very much.
It is preferably, described to generate original gamma-spectrometric data sample, specifically:
Original gamma-spectrometric data sample is generated using Monte Carlo simulation method.
Preferably, to after matrixing gamma-spectrometric data sample be standardized the formula of use are as follows:
Wherein,Value after indicating the jth road counting criteria of i-th of power spectrum, UijFor the jth of i-th of power spectrum
Road counts, and m and n indicate constant.
Preferably, the gamma-spectrometric data sample after described pair of standardization carries out zero averaging, the zero averaging formula are as follows:
Indicate average power spectrum,Indicate that the jth road of i-th of power spectrum counts the value after zero averaging, UijIt is i-th
The jth road of power spectrum counts, and m and n indicate constant.
Preferably, dimension-reduction treatment is carried out to the gamma-spectrometric data matrix being just distributed very much, to extract characteristics of energy spectrum vector
Feature, comprising:
One-dimensional vector in gamma-spectrometric data is expressed as two-dimensional matrix；
Feature extraction is carried out using sample set of the SVD method to gamma-spectrometric data, so that obtaining singular value features decomposes vector.
Preferably, described to carry out feature extraction using sample set of the SVD method to gamma-spectrometric data, to obtain singular value spy
Sign decomposes vector, and formula is as follows:
Ai=WiξiVi T,
AiRepresent each gamma spectrum two-dimensional matrix, WiRepresent the unitary matrice of a m*m, ξiRepresent the positive semidefinite pair of a m*n
Angular moment battle array, Vi TRepresent the associate matrix of a n*n.
Preferably, described to establish decision tree classification for described eigenvector feature as the input vector feature of disaggregated model
Device carries out in classification prediction the nucleic of gamma-spectrometric data:
Decision Tree algorithms use CART classification tree algorithm；
When doing classification prediction to the decision tree of generation, the sample of the test set of the gamma-spectrometric data is divided into the decision
After certain leaf node of tree, if there is multiple training samples in the leaf node, which takes the probability of the leaf node
Prediction classification of the maximum classification as the sample set.
Preferably, the classification predictablity rate of the nucleic that the gamma-spectrometric data is promoted using AdaBoost algorithm, packet
It includes:
Gather multiple decision tree classifiers and more wheel training are carried out to the nucleic data set in the gamma-spectrometric data；
The weight distribution that nucleic data set Radionuclide sample is corrected according to the error rate of single decision tree classifier, increases and divides
The weight of the nucleic sample of class mistake, while reducing the weight for correct nucleic sample of classifying；
According to the weight coefficient of decision tree classifier, the combination of corresponding decision Tree Classifier is formed.
Technical solution of the present invention has the advantages that
Technical solution of the present invention, by extracting characteristics of energy spectrum vector characteristics, and the characteristics of energy spectrum vector according to extraction is special
Sign establishes decision tree classifier and carries out classification prediction, then promotes the classification predictablity rate using AdaBoost algorithm.It answers
With AdaBoost algorithm promoted single-wheel decision tree classification result False Rate, thus greatly improve it is accurate to the classification of nucleic type
Rate.
Below by drawings and examples, technical scheme of the present invention will be described in further detail.
Detailed description of the invention
Fig. 1 is the flow chart of the recognition methods of the nucleic type of gamma-spectrometric data described in the embodiment of the present invention；
Fig. 2 is described in the embodiment of the present invention to the pretreated flow chart of gamma-spectrometric data；
Fig. 3 is the schematic diagram of the original gamma-spectrometric data form of mixing described in the embodiment of the present invention；
Fig. 4 is the schematic diagram of mixing nuclide identification precision described in the embodiment of the present invention；
Fig. 5 is the stream for extracting characteristics of energy spectrum vector characteristics described in the embodiment of the present invention in the gamma-spectrometric data being just distributed very much
Cheng Tu；
Fig. 6 is the process for promoting the classification predictablity rate described in the embodiment of the present invention using AdaBoost algorithm
Figure.
Specific embodiment
It will be appreciated that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Base
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts it is all its
Its embodiment, shall fall within the protection scope of the present invention.
The technical program is intended to take out orderly eigenmatrix data from power spectrum solution spectrum information, right by SVD method
Spectral information matrix carries out dimensionality reduction and extracts effective characteristic information.Effectively classification is carried out to feature using Decision Tree Algorithm to know
Not, repetition training takes turns recognition result more, the False Rate of single-wheel decision tree classification result is promoted using AdaBoost algorithm, thus greatly
The big classification accuracy improved to nucleic type.
In recent years, the extraction of spectral information data is processed, is abstracted into the number of orderly vector sum graphic form
According to carrying out spectrum unscrambling with specific learning algorithm, determine the classification of nucleic, belong to computer field in conjunction with industrial innovation, be
One emerging cross discipline.
The gamma-spectrometric data that sampling generates is interpreted as to the data of vector sum graphic form, is by original gamma-spectrometric data processing
Then matrix form carries out feature information extraction to energy spectrum matrix by SVD method, obtains characteristics of energy spectrum vector as subsequent mould
The training data of type differentiates that the feature of preceding ten dimension is just wrapped using decision Tree algorithms to ten dimensional features before extracting feature
The information characteristics of 90% power spectrum are contained, can effectively differentiate that the affiliated type of nucleic, the accuracy rate that single-wheel decision tree differentiates exist
73% or so, the lower recognition accuracy of single-wheel decision Tree algorithms is promoted using AdaBoost algorithm, by each round decision tree
Result of the classifier on training dataset is compared with practical true category result, according to the inclined of each round comparing result
Margin, the parameter of polyphony next round integral mould, to improve the accuracy of nuclide identification, final more wheel nuclide classification identifications are accurate
Rate is up to 98.33%.
The major technique that the present invention uses includes:
1. pair gamma-spectrometric data information extracts feature technology:
It is used to construct particle types, record power spectrum and definition decay using Monte Carlo simulation (Monte Carlo) method
Chain process generates multiple gamma-spectrometric data collection.11 groups of single nucleic and the mixing nucleic such as including Eu152, AM241, Na22, I131
Gamma-spectrometric data sample, the present invention carries out feature extraction processing to gamma-spectrometric data, using the road number of gamma spectrum as the dimension of vector,
Gamma-spectrometric data vector is extracted, one-dimensional vector is expressed as two-dimensional matrix form, is extracted using SVD method comprising most of
The singular value feature vector of gamma-spectrometric data information concentrates include can effectively differentiate nucleic type to excavate gamma-spectrometric data
Feature.
2. nucleic tagsort technology:
To being standardized per the numerical quantities grade for counting the feature vector formed together for each power spectrum, preceding ten are tieed up
Input sample feature of the feature vector, X 1 × 10 as decision-tree model, train the classification of nuclide identification, obtain single-wheel and determine
Plan tree algorithm is to nucleic type classification accuracy rate, using AdaBoost algorithm to the error rate of the single decision tree classifier of each round
The weight distribution of sample, increases the weight of the nucleic sample of classification error, while reducing classification constantly in amendment nucleic data set
The weight of correct nucleic sample, finally according to the weight coefficient of each round decision tree classifier, more the corresponding of wheel formation of formation are determined
The combination of plan Tree Classifier, obtains final nucleic category classification result.
Technical solution of the present invention is specific as shown in Figure 1, a kind of recognition methods of the nucleic type of gamma-spectrometric data, comprising:
Step S101: original gamma-spectrometric data is pre-processed, to obtain the gamma-spectrometric data matrix being just distributed very much；
Step S102: carrying out dimension-reduction treatment to the gamma-spectrometric data matrix that is just being distributed very much, thus extract characteristics of energy spectrum to
Measure feature；
Step S103: decision tree classifier is established using described eigenvector feature as the input vector feature of disaggregated model
Classification prediction is carried out to the nucleic of gamma-spectrometric data；
Step S104: the classification predictablity rate of the nucleic of the gamma-spectrometric data is promoted using AdaBoost algorithm.
In a preferred embodiment, described that original gamma-spectrometric data is pre-processed, to obtain the power spectrum being just distributed very much
Data, comprising:
Step S201: original gamma-spectrometric data sample is generated；
Step S202: to the original gamma-spectrometric data sample moment array；
Step S203: the gamma-spectrometric data sample after matrixing is standardized；
Step S204: zero averaging is carried out to the gamma-spectrometric data sample after standardization, to obtain the power spectrum being just distributed very much
Data matrix.
In a specific application scenarios, original gamma-spectrometric data is pre-processed specifically: use Monte Carlo simulation
(Monte Carlo) method generates original gamma-spectrometric data sample, each Monte Carlo gamma-spectrometric data collection regards a vector as,
Dimension of the road number of power spectrum as gamma spectrum data vector is taken, road location section is counted per energy together as every dimension
The value of feature.Z-Scorc standardization is carried out to multiple gamma-spectrometric data samples, mitigates the dimension impact of value between each dimension,
After standardizing gamma-spectrometric data, continue zero averaging, obtaining standard deviation is 1, the power spectrum that the standard that mean value is zero is just being distributed very much
Data.
Preferably, the formula of use is standardized to the gamma-spectrometric data sample after matrixing are as follows:
Wherein,Value after indicating the jth road counting criteria of i-th of power spectrum, UijFor the jth of i-th of power spectrum
Road counts, and m and n indicate constant.
Preferably, the gamma-spectrometric data sample after described pair of standardization carries out zero averaging, the zero averaging formula are as follows:
Indicate average power spectrum,Indicate that the jth road of i-th of power spectrum counts the value after zero averaging, UijIt is i-th
The jth road of power spectrum counts, and m and n indicate constant.
Monte Carlo simulation method:
Element identification needs with a large amount of sample to be to identify basis, obtains the feature comprising more comprehensively nucleic gamma-spectrometric data
Information, to improve nuclide identification accuracy.It is big to generate to be used herein as MNCP Monte Carlo simulation (Monte Carlo) method
Measure gamma-spectrometric data sample, including Eu152、AM241、Na22、I131Deng 11 groups of single nucleic and the gamma-spectrometric data sample of mixing nucleic.
Mixing gamma-spectrometric data form is shown in Fig. 3.
Fig. 3 is nucleic I131+Am241The original energy spectrum diagram of mixing, horizontal axis is energy road location section, and the longitudinal axis is that energy counts,
Each Monte Carlo gamma-spectrometric data collection regards a vector, dimension of the road number of gamma spectrum as vector, each power spectrum vector as
Are as follows:
Ui={ u1,u2,u3......un,
In formula, n ∈ { 1,1500 }, totally 1500 radio frequency channels.Each energy counts a dimension as vector, wherein i ∈
{ 1, m }, m are sample sizes.This symbiosis is at 1200 gamma-spectrometric data samples.
One-dimensional power spectrum vector is expressed as to the mode of two-dimensional characteristics of energy spectrum matrix, as follows:
One gamma-spectrometric data collection by m sample, each sample include n track data dimension.SVD carries out feature to sample set
It extracts to obtain singular value features and decompose vector ξi(i ∈ 1,2,3 ... m), the decline of the weights of singular value features it is very fast, it is unusual
Value matrix ξiPreceding t singular value include AiThe most information amount of matrix treats different nucleic and forms different singular value matrixs
Length, the preceding t singular value for uniformly extracting its singular value length constitute t dimension singular value vector, and enabling t is 10, extracts power spectrum
Energy more than 90%, obtains X1×10Feature vector, effective input feature value as training pattern.On this basis,
In conjunction with above-mentioned assorting process to gamma-spectrometric data Classification and Identification, higher recognition accuracy is obtained, mixing nuclide identification precision is such as
Fig. 4.
In a preferred embodiment, dimension-reduction treatment is carried out to the gamma-spectrometric data matrix being just distributed very much, to mention
Take characteristics of energy spectrum vector characteristics, comprising:
Step S501: the one-dimensional vector in gamma-spectrometric data is expressed as two-dimensional matrix；
Step S502: feature extraction is carried out using sample set of the SVD method to gamma-spectrometric data, to obtain singular value features
Decompose vector.
The one-dimensional vector of gamma spectrum set of data samples is expressed as two-dimensional matrix form, using SVD method to sample set
Carry out feature extraction come obtain singular value features decompose vector ξ i (i ∈ 1,2,3 ... m),
Feature extraction is carried out using sample set of the SVD method to gamma-spectrometric data, so that obtaining singular value features decomposes vector,
Formula is as follows:
Ai=WiξiVi T,
AiRepresent each gamma spectrum two-dimensional matrix, WiRepresent the unitary matrice of a m*m, ξiRepresent the positive semidefinite pair of a m*n
Angular moment battle array, Vi TRepresent the associate matrix of a n*n.Wherein, singular value matrix ξiPreceding t singular value include power spectrum AiSquare
The most information amount of battle array.It takes preceding 10 singular values of singular value length to constitute 10 dimension singular value vectors, obtains X1×10Feature
Input vector feature of the vector as disaggregated model.
In a preferred embodiment, described using described eigenvector feature as the input vector feature of disaggregated model
Decision tree classifier is established to carry out in classification prediction the nucleic of gamma-spectrometric data:
Decision Tree algorithms use CART classification tree algorithm；
When doing classification prediction to the decision tree of generation, the sample of the test set of the gamma-spectrometric data is divided into the decision
After certain leaf node of tree, if there is multiple training samples in the leaf node, which takes the leaf node
Prediction classification of the classification of maximum probability as the sample set.
Using categorised decision tree algorithm model, sum number in terms of the fitting effect on decision Tree algorithms nucleic data set it is predicted that
Aspect, which compares ridge regression algorithm and Lasso regression algorithm, has good classification to show, and can be used as Weak Classifier and further answer
The performance of lift scheme in AdaBoost Integrated Algorithm is used, and is capable of handling the numerical value that nucleic feature samples data are discrete value
Characteristic.Firstly, reading the nucleic feature vector of m group size, every feature vector matches actual nucleic label, forms data set
D(X(i),y(i))) do training.Decision Tree algorithms select characteristics of energy spectrum to establish decision tree using CART classification tree algorithm.Decision
It is as follows to set the tree node feature selecting formula established:
CART algorithm picks feature establishes decision tree nodes, using dichotomy, carries out two points to feature, wherein D table
Show power spectrum sample data set, D=(X(i),y(i)))i∈{1,m},y(i)∈ { 1,8 }, the characteristic attribute A=a of each power spectrumjj∈
{ 1, n }, sample class k ∈ { 1, K }, ajCharacteristic attribute value v ∈ { V }={ 1, V }, the optimum attributes a* of present node, and should
Corresponding optimum attributes value vbest, a* the value v ∈ { 1, V } of optimum attributes, wherein a*=argmin Gini (D, aj), { DV*}
Power spectrum sample data set corresponding to the property set A of removal optimum attributes a* is indicated, according to the corresponding optimum attributes of optimum attributes
Value, is divided into Sub Data Set { DV for data set DbestAnd { DV*, the y-bend tree node of present node is generated, and establish and work as prosthomere
The left and right node of point, left sibling corresponding data collection { DVbest, right node corresponding data collection { DV*, left and right node is reused
The tree node feature selecting formula that decision tree is established divides power spectrum sample data set, generates child node, generates decision tree.
Negated belonging to m group, one group of nucleic feature vector of same format is as test data set, in trained decision tree
On T, the accuracy rate of Self -adaptive Decision-Tree Classifier Model.Nucleic test set be accuracy of identification is about 73.33.Use decision tree
Training dataset is predicted, precision 91.43.
When giving a forecast to the decision tree of generation, after the sample of gamma-spectrometric data test set is divided into certain leaf node, if
There are multiple training samples in the leaf node, then the test sample takes the classification of the maximum probability of the leaf node as the sample
The prediction classification of this collection.
In a preferred embodiment, the classification of the nucleic that the gamma-spectrometric data is promoted using AdaBoost algorithm
Predictablity rate, comprising:
Step S601: gather multiple decision tree classifiers and more trainings in rotation are carried out to the nucleic data set in the gamma-spectrometric data
Practice；
Step S602: the weight point of nucleic data set Radionuclide sample is corrected according to the error rate of single decision tree classifier
Cloth, increases the weight of the nucleic sample of classification error, while reducing the weight for correct nucleic sample of classifying；
Step S603: according to the weight coefficient of decision tree classifier, the combination of corresponding decision Tree Classifier is formed.
According to decision tree classifier do classification prediction as a result, using Integrated Algorithm AdaBoost, gather multiple decision trees
Classifier carries out more wheel training to nucleic data set, and each round all constantly corrects core according to the error rate of single decision tree classifier
Prime number increases the weight of the nucleic sample of classification error, while reducing the correct nucleic sample of classification according to concentrating the weight of sample to be distributed
This weight, finally according to the weight coefficient of each round decision tree classifier, the corresponding decision Tree Classifier group that more wheels are formed
It closes, constructs the good strong classifier model of a Generalization Capability.Weighted error rate e1 formula of the first round decision tree classifier on D
It is as follows:
Wherein, f1(X(i)) represent first round decision tree classifier, data set D=(X(i),y(i)) i ∈ { 1, m }, y (i) ∈
{ 1,8 }, the sample weights distribution of first decision tree classifier, D (1)=(w11,w12,w13....w1m), w1i=1/m.
Calculate the weight coefficient α of first classifier1Formula:
α1=1/2 [log (1-e1)/e1],
Standardizing factor formula Zk:
Wherein, the sample set weight coefficient distributed update formula of the 2nd decision tree classifier is as follows:
W2i=[w1iexp(-αkI(f1(X(i)-y(i))>0.5))]/Z1,
Training the 2nd, 3... take turns decision tree classifier, count to K times until exercise wheel and stop, finally obtaining corresponding decision tree
Classifier corresponding weight coefficient α k and corresponding classifier fk(X(i)) as a result, Integrated Strategy takes turns category vote method using K, it obtains
To sample X(i)Final strong classifier F (X(i)) model, final training result improves to the extensive of gamma-spectrometric data identification prediction
Ability.
The present invention optimizes model parameter learning rate and training aids number using the method for the tune ginseng of cross validation,
The learning rate range of setting model is v:[0.1,1.0], it is n:[100 that model, which is iterated trained decision tree classifier number,
1000], joined by 5 folding cross validation tune to obtain accuracy rate of the model in test sample collection and show, finally obtain decision tree
The number of classifier is 300, and learning rate takes 0.8, and the classification prediction accuracy for obtaining nucleic test data set is 98.33%.
Technical solution of the present invention, in order to improve to power spectrum parsing to the ability of nucleic category identification, the invention proposes one
Kind of the Classification and Identification technology based on AdaBoost algorithm, relative to traditional power spectrum analytic method, the present invention is by the road of gamma spectrum
Dimension of the number as vector, extracts gamma-spectrometric data vector, one-dimensional power spectrum vector is expressed as two-dimensional characteristics of energy spectrum matrix, is adopted
With SVD method to gamma-spectrometric data dimension specification, feature is extracted to reduce the characteristic dimension of matrix, is extracted most bright in gamma-spectrometric data
Aobvious energy eigenvalue more effectively indicates that decision-tree model data input with more small data dimension, constructs AdaBoost model
The training result of more wheel CART decision classifying trees is combined using the combination strategy of K wheel category vote method, is promoted and is classified than single-wheel
The better classification results of model.Therefore, the gamma-spectrometric data collection combination machine learning algorithm by acquiring is attempted to carry out power spectrum number
According to solution spectrum analysis, better nuclide classification recognition effect is obtained, to being promoted the recognition accuracy of classification to more preferably imitating
Fruit.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
In the technical scope disclosed by the present invention, any changes or substitutions that can be easily thought of by those familiar with the art, all answers
It is included within the scope of the present invention.Therefore, protection scope of the present invention should be subject to the protection scope in claims.
Claims (10)
1. a kind of recognition methods of the nucleic type of gamma-spectrometric data characterized by comprising
Original gamma-spectrometric data is pre-processed, to obtain the gamma-spectrometric data matrix being just distributed very much；
Dimension-reduction treatment is carried out to the gamma-spectrometric data matrix being just distributed very much, to extract characteristics of energy spectrum vector characteristics；
Decision tree classifier is established to gamma-spectrometric data using described eigenvector feature as the input vector feature of disaggregated model
Nucleic carries out classification prediction；
The classification predictablity rate of the nucleic of the gamma-spectrometric data is promoted using AdaBoost algorithm.
2. the recognition methods of the nucleic type of gamma-spectrometric data according to claim 1, which is characterized in that described to be just distributed very much
Gamma-spectrometric data, specifically:
Standard deviation is 1, the gamma-spectrometric data that the standard that mean value is zero is just being distributed very much.
3. the recognition methods of the nucleic type of gamma-spectrometric data according to claim 2, which is characterized in that described to original energy
Modal data pretreatment, to obtain the gamma-spectrometric data being just distributed very much, comprising:
Generate original gamma-spectrometric data sample；
To the original gamma-spectrometric data sample moment array；
The gamma-spectrometric data sample after matrixing is standardized；
Zero averaging is carried out to the gamma-spectrometric data sample after standardization, to obtain the gamma-spectrometric data matrix being just distributed very much.
4. the recognition methods of the nucleic type of gamma-spectrometric data according to claim 3, which is characterized in that the generation is original
Gamma-spectrometric data sample, specifically:
Original gamma-spectrometric data sample is generated using Monte Carlo simulation method.
5. the recognition methods of the nucleic type of gamma-spectrometric data according to claim 3, which is characterized in that after matrixing
Gamma-spectrometric data sample be standardized the formula of use are as follows:
Wherein,Value after indicating the jth road counting criteria of i-th of power spectrum, UijFor the jth road meter of i-th of power spectrum
Number, m and n indicate constant.
6. the recognition methods of the nucleic type of gamma-spectrometric data according to claim 3, which is characterized in that described pair of standardization
Gamma-spectrometric data sample afterwards carries out zero averaging, the zero averaging formula are as follows:
Indicate average power spectrum,Indicate that the jth road of i-th of power spectrum counts the value after zero averaging, UijFor i-th power spectrum
Jth road counts, and m and n indicate constant.
7. the recognition methods of the nucleic type of gamma-spectrometric data according to claim 1, which is characterized in that just dividing very much described
The gamma-spectrometric data matrix of cloth carries out dimension-reduction treatment, to extract characteristics of energy spectrum vector characteristics, comprising:
One-dimensional vector in gamma-spectrometric data is expressed as two-dimensional matrix；
Feature extraction is carried out using sample set of the SVD method to gamma-spectrometric data, so that obtaining singular value features decomposes vector.
8. the recognition methods of the nucleic type of gamma-spectrometric data according to claim 7, which is characterized in that described to use SVD
Method carries out feature extraction to the sample set of gamma-spectrometric data, so that obtaining singular value features decomposes vector, formula is as follows:
Ai=WiξiVi T,
AiRepresent each gamma spectrum two-dimensional matrix, WiRepresent the unitary matrice of a m*m, ξiThe positive semidefinite of a m*n is represented to angular moment
Battle array, Vi TRepresent the associate matrix of a n*n.
9. the recognition methods of the nucleic type of gamma-spectrometric data according to claim 1, which is characterized in that described by the spy
Sign vector characteristics establish decision tree classifier as the input vector feature of disaggregated model and classify to the nucleic of gamma-spectrometric data
In prediction:
Decision Tree algorithms use CART classification tree algorithm；
When doing classification prediction to the decision tree of generation, the sample of the test set of the gamma-spectrometric data is divided into the decision tree
After certain leaf node, if there is multiple training samples in the leaf node, which takes the maximum probability of the leaf node
Prediction classification of the classification as the sample set.
10. the recognition methods of the nucleic type of gamma-spectrometric data according to claim 1, which is characterized in that the use
AdaBoost algorithm promotes the classification predictablity rate of the nucleic of the gamma-spectrometric data, comprising:
Gather multiple decision tree classifiers and more wheel training are carried out to the nucleic data set in the gamma-spectrometric data；
The weight distribution that nucleic data set Radionuclide sample is corrected according to the error rate of single decision tree classifier, it is wrong to increase classification
The weight of nucleic sample accidentally, while reducing the weight for correct nucleic sample of classifying；
According to the weight coefficient of decision tree classifier, the combination of corresponding decision Tree Classifier is formed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811312752.6A CN109635650A (en) | 2018-11-06 | 2018-11-06 | The recognition methods of the nucleic type of gamma-spectrometric data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811312752.6A CN109635650A (en) | 2018-11-06 | 2018-11-06 | The recognition methods of the nucleic type of gamma-spectrometric data |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109635650A true CN109635650A (en) | 2019-04-16 |
Family
ID=66067376
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811312752.6A Pending CN109635650A (en) | 2018-11-06 | 2018-11-06 | The recognition methods of the nucleic type of gamma-spectrometric data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109635650A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111045072A (en) * | 2019-12-27 | 2020-04-21 | 核工业北京地质研究院 | Is suitable for CeBr3Gamma-energy spectrum iterative spectrum-solving algorithm of detector |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102313897A (en) * | 2010-06-29 | 2012-01-11 | 成都理工大学 | Radioactive spectrum identification method |
| CN102819745A (en) * | 2012-07-04 | 2012-12-12 | 杭州电子科技大学 | Hyper-spectral remote sensing image classifying method based on AdaBoost |
| CN107272048A (en) * | 2017-07-07 | 2017-10-20 | 西南科技大学 | A kind of complicated nuclide identification method based on fuzzy decision-tree |
| CN107390259A (en) * | 2017-07-14 | 2017-11-24 | 西南科技大学 | A kind of nuclide identification method based on SVD and SVM |
-
2018
- 2018-11-06 CN CN201811312752.6A patent/CN109635650A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102313897A (en) * | 2010-06-29 | 2012-01-11 | 成都理工大学 | Radioactive spectrum identification method |
| CN102819745A (en) * | 2012-07-04 | 2012-12-12 | 杭州电子科技大学 | Hyper-spectral remote sensing image classifying method based on AdaBoost |
| CN107272048A (en) * | 2017-07-07 | 2017-10-20 | 西南科技大学 | A kind of complicated nuclide identification method based on fuzzy decision-tree |
| CN107390259A (en) * | 2017-07-14 | 2017-11-24 | 西南科技大学 | A kind of nuclide identification method based on SVD and SVM |
Non-Patent Citations (4)
| Title |
|---|
| ABC_138: "机器学习：AdaBoost算法", 《CSDN-HTTPS://BLOG.CSDN.NET/ABC_138/ARTICLE/DETAILS/82720798》 * |
| S. FORKAPIC等: "orrelation analysis of the natural radionuclides in soil and indoor radon in Vojvodina, Province of Serbia", 《JOURNAL OF ENVIRONMENTAL RADIOACTIVITY》 * |
| 刘灏霖 等: "基于K-SVD的γ能谱特征提取及核素识别", 《电脑知识与技术》 * |
| 码农JAKE: "z-score 标准化(zero-mean normalization)", 《知乎-HTTPS://ZHUANLAN.ZHIHU.COM/P/32482328》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111045072A (en) * | 2019-12-27 | 2020-04-21 | 核工业北京地质研究院 | Is suitable for CeBr3Gamma-energy spectrum iterative spectrum-solving algorithm of detector |
| CN111045072B (en) * | 2019-12-27 | 2023-08-15 | 核工业北京地质研究院 | Be applicable to CeBr 3 Gamma energy spectrum iteration spectrum solving method of detector |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107451614B (en) | Hyperspectral classification method based on fusion of space coordinates and space spectrum features | |
| WO2018014610A1 (en) | C4.5 decision tree algorithm-based specific user mining system and method therefor | |
| CN103632168A (en) | Classifier integration method for machine learning | |
| CN102324038B (en) | Plant species identification method based on digital image | |
| CN107563442B (en) | Hyperspectral image classification method based on sparse low-rank regular graph tensor embedding | |
| CN108540451A (en) | A method of classification and Detection being carried out to attack with machine learning techniques | |
| CN108171136A (en) | A kind of multitask bayonet vehicle is to scheme to search the system and method for figure | |
| CN101256631B (en) | Method and apparatus for character recognition | |
| CN102571486A (en) | Traffic identification method based on bag of word (BOW) model and statistic features | |
| CN110135167A (en) | A kind of edge calculations terminal security grade appraisal procedure of random forest | |
| CN102982338A (en) | Polarization synthetic aperture radar (SAR) image classification method based on spectral clustering | |
| CN103838744A (en) | Method and device for analyzing query requirement | |
| CN108985360A (en) | Hyperspectral classification method based on expanding morphology and Active Learning | |
| CN111507385B (en) | Extensible network attack behavior classification method | |
| CN101833667A (en) | Pattern recognition classification method expressed based on grouping sparsity | |
| CN112149758A (en) | Hyperspectral open set classification method based on Euclidean distance and deep learning | |
| CN109213853A (en) | A kind of Chinese community's question and answer cross-module state search method based on CCA algorithm | |
| CN107390259A (en) | A kind of nuclide identification method based on SVD and SVM | |
| CN104166691A (en) | Extreme learning machine classifying method based on waveform addition cuckoo optimization | |
| CN112766161B (en) | Hyperspectral target detection method based on integrated constraint multi-example learning | |
| CN104142960A (en) | Internet data analysis system | |
| CN104318515A (en) | Hyper-spectral image wave band dimension descending method based on NNIA evolutionary algorithm | |
| CN106911591A (en) | The sorting technique and system of network traffics | |
| CN114266961A (en) | Method for integrating, learning and classifying marsh vegetation stacks by integrating hyperspectral and multiband fully-polarized SAR images | |
| CN104751184A (en) | Fully polarimetric SAR image classification method based on sparse strength statistics |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190416 |