CN108491923A - Based on the pest image-recognizing method for improving wolf pack algorithm optimization Bayesian network - Google Patents
Based on the pest image-recognizing method for improving wolf pack algorithm optimization Bayesian network Download PDFInfo
- Publication number
- CN108491923A CN108491923A CN201810316604.5A CN201810316604A CN108491923A CN 108491923 A CN108491923 A CN 108491923A CN 201810316604 A CN201810316604 A CN 201810316604A CN 108491923 A CN108491923 A CN 108491923A
- Authority
- CN
- China
- Prior art keywords
- wolf
- bayesian network
- algorithm
- wolf pack
- pest
- 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
- 241000282461 Canis lupus Species 0.000 title claims abstract description 126
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 71
- 241000607479 Yersinia pestis Species 0.000 title claims abstract description 49
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000005457 optimization Methods 0.000 title claims abstract description 18
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 23
- 230000006399 behavior Effects 0.000 claims abstract description 20
- 238000012549 training Methods 0.000 claims abstract description 20
- 229910002056 binary alloy Inorganic materials 0.000 claims abstract description 15
- 238000000605 extraction Methods 0.000 claims abstract description 15
- 238000012360 testing method Methods 0.000 claims abstract description 14
- 230000005012 migration Effects 0.000 claims abstract description 11
- 238000013508 migration Methods 0.000 claims abstract description 11
- 230000003993 interaction Effects 0.000 claims abstract description 9
- 230000035772 mutation Effects 0.000 claims abstract description 8
- 230000008569 process Effects 0.000 claims abstract description 8
- 230000000739 chaotic effect Effects 0.000 claims abstract description 6
- 238000010276 construction Methods 0.000 claims abstract description 4
- 238000012545 processing Methods 0.000 claims abstract description 4
- 230000006872 improvement Effects 0.000 claims abstract description 3
- 238000013459 approach Methods 0.000 claims abstract 2
- 230000006870 function Effects 0.000 claims description 12
- 238000001514 detection method Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000011161 development Methods 0.000 claims description 5
- 230000008901 benefit Effects 0.000 claims description 3
- 238000013461 design Methods 0.000 claims description 3
- 235000015170 shellfish Nutrition 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 claims 1
- 238000007635 classification algorithm Methods 0.000 claims 1
- 238000007476 Maximum Likelihood Methods 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 6
- 235000013305 food Nutrition 0.000 description 5
- 241000282421 Canidae Species 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 201000010099 disease Diseases 0.000 description 2
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 239000000575 pesticide Substances 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000004083 survival effect Effects 0.000 description 2
- 201000004569 Blindness Diseases 0.000 description 1
- 241001481710 Cerambycidae Species 0.000 description 1
- 241000254173 Coleoptera Species 0.000 description 1
- 241000238631 Hexapoda Species 0.000 description 1
- 241000254043 Melolonthinae Species 0.000 description 1
- 241001062472 Stokellia anisodon Species 0.000 description 1
- 238000012271 agricultural production Methods 0.000 description 1
- 230000009418 agronomic effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000009432 framing Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 230000009191 jumping Effects 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 230000001568 sexual effect Effects 0.000 description 1
- 239000002689 soil Substances 0.000 description 1
- 241000894007 species Species 0.000 description 1
- 238000003045 statistical classification method Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
- G06F18/24155—Bayesian classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Molecular Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The present invention proposes a kind of pest image-recognizing method based on improvement wolf pack algorithm optimization Bayesian network, and this approach includes the following steps:Step 1:It is proposed a kind of improved binary system wolf pack algorithm (Improved binary Wolf Colony Algorithm, IBWCA), wolf pack algorithm is improved, mutation operator is added in the migration behavior for visiting wolf, Approximation Operator is added in calling behavior, and interaction operator is added in jointly attack behavior, the mode of proposition chaotic motion updates wolf pack in the renewal process of wolf pack;Step 2:Bayesian Structure Learning optimization algorithm (Bayesian Network Construction algorithm using IBWCA, BNC IBWCA) based on improved binary system wolf pack algorithm;Step 3:The identifying processing of pest image is carried out in conjunction with convolutional neural networks and Bayesian network.Feature extraction is carried out to the picture of training set and test set with pre-training good convolutional neural networks, Classification and Identification is carried out with Bayes classifier.
Description
Technical field
The present invention relates to convolutional neural networks, Bayesian network and wolf pack algorithm.
Background technology
The problem of diseases and pests of agronomic crop, is closely bound up with people’s lives, directly affects the development of agricultural.At present to evil
The most common method that worm is prevented and administered is then to use pesticide.However, various due to pest species, people are lacking
In the case of identification, blindness uses pesticide, does not suit the remedy to the case, does not simply fail to effectively administer pest, also tends to destroy soil
Earth and environment influence the sustainable development of agricultural production, make the output of crops that the trend of negative growth be presented.The morphologic species of pest
Class is various, and the identification of traditional pest only leans on naked eyes to be identified to observe the feature of pest, there are certain subjectivity with
Limitation, and working strength is very big, it is time-consuming and laborious.Therefore, how carry out pest identification rapidly and efficiently, which becomes, is administered disease pest
The vital step of evil.
The identification technology of pest be influence the fast-developing important link of agricultural, and realize agricultural modernization it is crucial because
Element has had many since the defect of artificial pest identification has been difficult the requirement for height for meeting people and being identified for pest
Scholar proposes many novel pests with various advanced technologies and knows method for distinguishing, for example identified by sound, passes through
Context awareness passes through image recognition.Most effective one is identified by the image of pest, and the pest based on image is known
Other technology, is currently a new research field of comparison, and many experts and scholars both domestic and external are also studying.Current evil
Worm identification technology has certain limitation:(1) traditional Insect Pest Identification is point feature and line feature based on image,
In the case where illumination condition is complicated or photo angle is changeable, recognition effect is bad.(2) traditional image-recognizing method is only
It is only extracted feature representative in input picture, there is certain limitation.
Currently, machine vision technique in the identification technology of pest, there is prodigious development prospect.Bayesian network is as machine
Device study one, can use graph theory the direct showing problem of language structure and according to probability theory principle to problem
Structure carries out analysis and utilization, reduces the complexity of reasoning.Therefore, Bayes classifier is calculated on classifying quality compared with other classification
Method has metastable classifying quality.But the Structure learning of Bayesian network is always the problem of a NP hardly possible.Herein we
Use the method based on search and scoring.It proposes a kind of improved wolf pack algorithm, the more new strategy of wolf pack is designed, to improve part
Extreme-value problem improves learning algorithm precision, makes the Structure learning it is suitable for Bayesian network, by bayesian network structure
Habit process is converted into the problem of finding optimal head wolf.Then in conjunction with convolutional neural networks advantage exclusive in terms of feature extraction,
Pest is identified.
In conclusion it is proposed that a kind of based on the pest image recognition side for improving wolf pack algorithm optimization Bayesian network
Method carries out the positioning of pest target with the salient region detection method based on global contrast first, then to pest target
Region is divided using GrabCut algorithms, is saved in data file automatically, formed data set, for model training and
Test.And then with the good convolutional neural networks of pre-training, the characteristics of image on training set and test set is extracted, inputs Bayes
Network.Then traditional wolf pack algorithm is improved, as searching algorithm, bayesian information criterion (Bayesian
Information Criterion, BIC) it is used as score function, learn the structure of Bayesian network.Then maximum likelihood is used again
(Maximum Likelihood, ML) algorithm learns the parameter of Bayesian network, forms Bayes classifier.
Invention content:
To solve the problems, such as agricultural pest, set forth herein a kind of based on the pest for improving wolf pack algorithm optimization Bayesian network
Image-recognizing method.INVENTION IN GENERAL includes:It is proposed a kind of improved binary system wolf pack algorithm.With improved binary system wolf pack
Algorithm carries out the Structure learning of Bayesian network as searching algorithm, forms Bayes classifier.In conjunction with convolutional neural networks
The classification feature of feature extraction functions and Bayes classifier carries out Classification and Identification to pest image.
A kind of pest image-recognizing method based on improvement wolf pack algorithm optimization Bayesian network, it is characterised in that:At least
Include the following steps:
Step 1:It is proposed a kind of improved binary system wolf pack algorithm (Improved binary Wolf Colony
Algorithm, IBWCA)
Wolf pack algorithm is improved, mutation operator is added in the migration behavior for visiting wolf, is added and forces in calling behavior
Nearly operator, and interaction operator is added in jointly attack behavior.And in the update step of wolf pack, the side using chaotic maps is proposed
Formula generates new artificial wolf and replaces superseded artificial wolf.
Step 2:Bayesian Structure Learning optimization algorithm (Bayesian based on improved binary system wolf pack algorithm
Network Construction algorithm using IBWCA,BNC-IBWCA)
Bayesian network structure can indicate that corresponding structure matrix is encoded to { x with adjacency matrix11,, x12...,
x1n, x21,, x22..., x2n..., xn1,, xn2..., xnn}.Then it is being used as search with improved binary system wolf pack algorithm
Algorithm, BIC seek optimal bayesian network structure as score function.
Step 3:The identifying processing of pest image is carried out in conjunction with convolutional neural networks and Bayesian network
Feature extraction is carried out to the picture of training set and test set with pre-training good convolutional neural networks, inputs training set
The characteristic attribute of upper extraction and classification carry out the study of bayesian network structure with BNC-IBWCA, then use ML to carry out shellfish again
The study of this network parameter of leaf forms one and the most matched Bayesian network of input data set, as Bayes's classification
Device.By the characteristic attribute extracted on test set and classification input Bayes classifier, Bayes classifier is tested.
Advantageous effect:
Compared with prior art, using design scheme of the present invention, following technique effect can be reached:
1, improved binary system wolf pack algorithm is more suitable for carrying out Bayesian network knot relative to traditional wolf pack algorithm
The study of structure, in the walk process for visiting wolf, the mutation operator of addition ensure that the randomness for visiting wolf migration, effectively avoid
Algorithm is absorbed in locally optimal solution, and in calling behavior, the Approximation Operator of addition is effectively guaranteed overall towards global optimum
The direction development and change of solution, the interaction operator being added in jointly attack behavior, are effectively guaranteed the interaction of information between wolf pack individual.
In the update step of wolf pack, in the way of chaotic maps in the way of generate new artificial wolf, ensure that the diversity of population.
2, the method for the bayesian network structure learning based on wolf pack algorithm, relative to traditional bayesian network structure
Learning method carries out shellfish using improved binary system wolf pack algorithm as searching algorithm using BIC scorings as score function
Ye Sisi network structures learn, and the process of the Structure learning of Bayesian network is converted into the scoring optimizing of corresponding adjacency matrix
Problem.Based on wolf pack algorithm optimization Bayesian Structure Learning method is improved, it is effectively controlled search precision and convergence rate, it can
To avoid being absorbed in locally optimal solution, and robustness is calculated with preferable, ensure that the accurate of bayesian network structure learning
Degree.
3, the pest image recognition of complex background is carried out in conjunction with conspicuousness detection, convolutional neural networks and Bayesian network,
Salient region detection method based on global contrast carries out the pest in picture the positioning of pest target, then to pest mesh
Target area is divided automatically using GrabCut algorithms, can be effective using the image handled well as raw data set
The complex background for solving the problems, such as the image of pest, further promotes discrimination.Then by the good volume of image data input pre-training
Product neural network carries out feature extraction, then using the feature vector of extraction as input, training and test Bayes classifier.Due to
Convolutional neural networks do not need complicated pretreatment work early period, can directly input image, but sorter makees completely by complete
Articulamentum carries out, and classifying quality is bad, and Bayes classifier is compared with other graders, its design method be it is a kind of most
Basic statistical classification method, classifying quality is more accurate, but is only manually operated to the extraction of feature, very inaccurate
Really.Therefore the present invention has drawn convolutional neural networks and the advantage of Bayesian network respectively, keeps away its disadvantage, on classification capacity
Prodigious promotion is arrived.
Description of the drawings:
Disaggregated model structural framing figures of Fig. 1 based on Bayesian network and convolutional neural networks
The improved wolf pack algorithm flow charts of Fig. 2
Fig. 3 Bayes classifier learning process figures
Specific implementation mode:
Step 1:It is proposed a kind of improved binary system wolf pack algorithm (Improved binary Wolf Colony
Algorithm, IBWCA)
It is assumed that search space is m*m dimensions, N indicates the number of artificial wolf in wolf pack, and the position of i-th of artificial wolf is defined
For Xi={ X11,...,X1m,X21,...X2m,...,Xm1,...,Xmm}.Each artificial wolf represents a feasible solution, his what is heard
The size of the food concentration Y arrived represents the good and bad degree of solution.
Step 1:Numerical value initializes.
(1) number N and position X of wolf pack are initializedi
If completely random initializes the position X of wolf packiWith number N can give algorithm increase searching times, so, here I
The position of artificial wolf in wolf pack is initialized using Mutual Information Theory, although Mutual Information Theory such as method judges set membership,
It is that whether there is dependence between may determine that 2 variables, finds out point there are dependence with it, may deposits between them
On side, then N wolf and its position X are generated according to these dependencesi。
(2) maximum iteration Kmax
(3) maximum migration number Tmax
(4) the number Z of worst wolf
(5) migration step-length stepa, long-range raid step-length stepb, besiege step-length stepc(step-length is integer, indicates search essence
Degree)
(6) direction of search h
Step 2:The generation and migration behavior of head wolf
Define 1:Mutation operator Θ=(Xi, random (stapa)):Wherein, XiIndicate the position encoded X of spy wolf ii=
{Xi1... .Xim 2, random (stapa) indicate visit wolf it is position encoded in randomly select stapaA position carries out coding
Mutation (is changed to 0 if being encoded to 1, is changed to 1) if being encoded to 0.
It is head wolf to choose optimal artificial wolf, and all artificial wolves in addition to head wolf are to visit wolf, and execute migration behavior, if Yi
>Ylead, then visit wolf i and substitute head wolf and initiate calling behavior;If Yi<=YleadIt then visits wolf to take a step forward to h direction respectively, walk
A length of stepa, i.e., to visit wolf execute h mutation operator Θ, record to all directions take a step forward after food concentration YNewly, choose
Wherein maximum value YNew maxIf YNew max>Yi, then Y is usedNew maxTo deserved position XNewlyReplace original position Xi, repeat more than migration
Behavior, until Yi>YleadOr migration number reaches Tmax。
Step 3:Calling behavior
Define 2:Approximation Operator Ψ=(Xhead, Xm, random (stepb)):Wherein XheadIndicate position encoded, the X of head wolfm
Indicate the position encoded of violent wolf, random (stepb) indicate that it is step that one section of continuous length is randomly selected in head wolfbPosition
It sets, replaces the coding at same position in violent wolf.
Head wolf calls violent wolf constantly to be approached to the position of head wolf, wherein it is violent wolf to set out whole wolves outside a wolf.Violent wolf with
Larger step-length stepbIt is approached to head wolf, that is, violent wolf is allowed to execute Approximation Operator Ψ, such as Xhead(100101100), Xm
(100010101) 2-6 replace violent wolf 2-6 in head wolf are chosen, X is obtainedM is new(100101101), step-length is 5, this
Version embodies influence and guidance of the globally optimal solution to individual.During being approached to head wolf, if Ym>Yhead, then should
Violent wolf is updated to a wolf and initiates new round calling behavior;If Ym<=Yhead, then violent wolf i continue the direction approximation towards head wolf,
Until being less than judgement distance d at a distance from violent wolf is between head wolfnear。
Step 4:Jointly attack behavior
Define 3:Interaction operator Δ=(Xy,Xz,random(stepc)):Wherein XyAnd XzIt indicates to besiege two in range
Artificial wolf, random (stepc), it is step that expression, which randomly selects continuous length in an artificial wolf,cIt is position encoded, and it is another
The coding of one artificial wolf same position is interchangeable.
Under the commander of head wolf, violent wolf and spy wolf besiege prey, herein, it will be assumed that, the position of head wolf
XheadEven if the position X of preyfood, participate in besiege artificial wolf distance all in the region of a very little, between them away from
It is close from very, it in order to capture food as soon as possible, needs to carry out information sharing between each other, that is, executes interaction operator Δ, calculate
Prey concentration Y, if replaced YnewMore than the Y before replacementold, then compiled with the position before replaced position encoded replacement
Code is otherwise, position encoded constant.
Step 5:Update wolf pack
Because food is preferentially given strong wolf by the survival mechanism of " survival of the fittest ", then small and weak wolf will be starved
Extremely, we are updated wolf pack, eliminate poor Z only artificial wolves, then randomly generate Z only artificial wolves, such update mechanism,
It can keep the diversity of population and effectively avoid limiting into locally optimal solution.
Chaos is to determine the intrinsic stochasticity of sexual system, is not repeatedly traversed by itself " rule " within the limits prescribed
The randomness of all patents, chaos will not be reduced relative to common with the increase of information content at random, therefore, Wo Menli
With the mode of chaotic maps, the small and weak wolf for eliminating is replaced to generate new Z wolf, makes the newly-generated population be not in
The individual repeated, and then improve the diversity of population.
Step 6:Judge whether to terminate
When the result acquired meets the requirements or reached maximum iteration KmaxWhen, terminate the algorithm, export head
The position of wolf.Otherwise jumping to step 2.
Step 2:Bayesian Structure Learning optimization algorithm (Bayesian Network based on improved wolf pack algorithm
Construction algorithm using IBWCA,BNC-IBWCA)
1. this paper bayesian network structure learning coding modes
In the bayesian network structure of a n node, bayesian network structure can use an adjacency matrix X=
{xijIndicate, whereinIts corresponding structure matrix is encoded to { x11,, x12..., x1n, x21,,
x22..., x2n..., xn1,, xn2..., xnn}.Both it had been that bayesian network structure encodes.
2. score function
We select bayesian information criterion herein, and abbreviation BIC scorings, it is under the premise of large sample to edge likelihood
A kind of approximation of function.
If G is one by n variable X={ X1, X2... XnComposition a bayesian network structure, each variable
XiThere is riA value 1,2 ... ri, XiFather node Xpa(i) value can have qiKind, it is based on Bayesian network G and data set
Quality between Q can be measured with BIC score functions.
The formula first item indicates that the fitting degree of structure and data, Section 2 are to penalize item about the complexity of model, keeps away
Exempt from overfitting.
Score function value is bigger, indicates that performance is better.
3. searching algorithm
IBWCA is used to carry out the study of bayesian network structure as searching algorithm herein.
It is assumed that search space is m*m dimensions, N indicates that the number of wolf in wolf pack, m indicate the section in bayesian network structure
It counts out, the position of i-th of artificial wolf is defined as Xi={ X11,...,X1m,X21,...X2m,...,Xm1,...,Xmm}.Each
Artificial wolf represents a feasible Bayesian network, the size of the food concentration Y that he is smelt represent bayesian network structure with
The fitting degree of test data, the position where algorithm terminates back wolf is exactly best bayesian network structure.Artificial wolf
It is a directed acyclic graph that the corresponding bayesian network structure figure in position, which will meet,.In the process of bayesian network structure learning
In, constantly edged is deleted when changing, and may lead to a cycle occur, and generate trivial solution in an iterative process, to destroy
The robustness of coding.Therefore, when being changed every time to wolf pack position encoded, it will judge that the artificial wolf after changing is corresponding
Bayesian network structure whether there is ring, judge whether there is ring with depth-priority-searching method (DFS) herein, if any, give up
The position re-starts operation.
Step 3:The identifying processing of pest image is carried out in conjunction with convolutional neural networks and Bayesian network
Step 1:Feature extraction is carried out to the picture of training set and test set with pre-training good convolutional neural networks
Herein for beetle, chafer, locust, longicorn, the common pest progress Classification and Identification of 40 class such as moth, initial data
It is obtained by manually shooting to collect and search for two ways by search engine, then original image is screened, chooses 9000
Pictures treat identical pest, and different stages of growth, by simple marking, they are two sons of adult and larva of same pest
Class handles and (is simply classified as two classes), then carries out pest with the salient region detection method based on global contrast
Then the positioning of target is divided using GrabCut algorithms, is then saved in then again to pest mesh target area automatically
In data file, data set is formed, is used for the training and test of model.
The good convolutional neural networks of pre-training on large data sets (pest data set) are chosen, it will be two-dimentional in data set
Picture is input to directly as input in convolutional neural networks model, then alternately by several volume bases and pond layer,
Feature of the image in different aspect in extraction.
Step 2:By the characteristic attribute extracted on training set and training sample input study Bayes classifier
(1) study of bayesian network structure
The study of bayesian network structure is carried out with BNC-IBWCA proposed above
(2) study of Bayesian network parameters
Bayesian network parameters study is not the research emphasis of this patent, we simply carry out parametrics with ML algorithms
It practises.
Maximum likelihood (ML) estimation is based on traditional statistical analysis, he judges the quasi- of sample and model according to likelihood score
Conjunction degree, likelihood function are as shown in formula 1.
Maximal possibility estimation is exactly the strictly maximum Θ of likelihood function.
Θ=argmaxΘL (Θ, D) formula 2
Step 4:The characteristic attribute extracted on test set and sample classification are inputted into Bayes classifier, to grader
It is tested.
Test data set is input in the Bayes classifier succeeded in school, predicts its classification, checks the classification of grader
Effect then is continuing to improve if there is deviation.
Claims (4)
1. a kind of based on the pest image-recognizing method for improving wolf pack algorithm optimization Bayesian network, it is characterised in that:At least wrap
Include following steps:
Step 1:It is proposed a kind of improved binary system wolf pack algorithm (Improved binary Wolf Colony
Algorithm, IBWCA)
Wolf pack algorithm is improved, mutation operator is added in the migration behavior for visiting wolf, is added in calling behavior and approaches calculation
Son, and interaction operator is added in jointly attack behavior, and in the update step of wolf pack, chaotic motion is added, ensure that population
Diversity;
Step 2:Bayesian network structure learning optimization algorithm (Bayesian based on improved binary system wolf pack algorithm
Network Construction algorithm using IBWCA,BNC-IBWCA)
Improved wolf pack algorithm is applied to the study of bayesian network structure, is used as and searches with improved binary system wolf pack algorithm
Rope algorithm, BIC are score function, seek optimal bayesian network structure;
Step 3:The identifying processing of pest image is carried out in conjunction with convolutional neural networks and Bayesian network
Feature extraction is carried out to the picture of training set and test set with pre-training good convolutional neural networks, input training set above carries
The characteristic attribute taken and classification carry out the Structure learning of Bayesian network with BNC-IBWCA, then use ML algorithms to carry out shellfish again
The study of the parameter of this network of leaf forms one and the most matched Bayesian network of input data set, as Bayes point
Class device tests Bayes classifier the characteristic attribute extracted on test set and classification input Bayes classifier.
2. according to claim 1 a kind of based on the pest image recognition side for improving wolf pack algorithm optimization Bayesian network
Method, it is characterised in that:The improved binary system wolf pack algorithm proposed in step 1, make it is suitable for solve bayesian network structure
Study, visit wolf walk process in, it is proposed that mutation operator, ensure that visit wolf migration randomness, effectively avoid
It is absorbed in locally optimal solution;In calling behavior, it is proposed that Approximation Operator keeps every wolf close to optimal wolf with certain rate,
It is effectively guaranteed the overall direction development and change towards globally optimal solution;The interaction operator proposed in jointly attack behavior, then effectively
The interaction that ensure that information between wolf pack individual, so that its wolf pack individual is gradually moved to preferably position;In the update of wolf pack
In step, proposition generates new wolf to replace superseded wolf in the way of chaotic maps, ensure that the diversity of population.
3. according to claim 1 a kind of based on the pest image recognition side for improving wolf pack algorithm optimization Bayesian network
Method, it is characterised in that:Using improved binary system wolf pack algorithm as searching algorithm in step 2, using BIC scorings as commenting
Divide function, carries out bayesian network structure learning, the process of bayesian network structure learning is converted into corresponding adjacency matrix
Scoring optimization problem, based on improve wolf pack algorithm optimization bayesian network structure learning method, be effectively controlled search
Precision and convergence rate, can be to avoid being absorbed in locally optimal solution, and calculates robustness with preferable, ensure that Bayesian network
The accuracy of network Structure learning.
4. according to claim 1 a kind of based on the pest image recognition side for improving wolf pack algorithm optimization Bayesian network
Method, it is characterised in that:Conspicuousness detection, convolutional neural networks and Bayesian network is combined to carry out the evil of complex background in step 3
Convolutional neural networks are only used for feature extraction by worm image recognition, using the feature vector of extraction as input, use Bayes's classification
Algorithm is classified, and makees feature extraction with the convolutional neural networks of pre-training, with improvement wolf pack algorithm optimization Bayesian Structure
The Bayesian network obtained is practised as grader, is organically combined, is learnt from other's strong points to offset one's weaknesses by the two, performance convolutional neural networks can be integrated
With Bayesian network in terms of feature extraction and classification the advantages of, it is unstable and pass in turn avoid convolutional neural networks classifying quality
The shortcomings of artificial design features extracting method adaptivity of uniting is poor is based on to be greatly improved on classification capacity
The salient region detection method of global contrast carries out the positioning of pest target, then to pest mesh target area, uses
GrabCut algorithms are divided automatically, are capable of the complex background problem of the image of effective solution pest, further promote identification
Rate.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810316604.5A CN108491923B (en) | 2018-04-10 | 2018-04-10 | Pest image recognition method based on improved wolf colony algorithm optimized Bayesian network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810316604.5A CN108491923B (en) | 2018-04-10 | 2018-04-10 | Pest image recognition method based on improved wolf colony algorithm optimized Bayesian network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108491923A true CN108491923A (en) | 2018-09-04 |
CN108491923B CN108491923B (en) | 2021-09-24 |
Family
ID=63315467
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810316604.5A Expired - Fee Related CN108491923B (en) | 2018-04-10 | 2018-04-10 | Pest image recognition method based on improved wolf colony algorithm optimized Bayesian network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108491923B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111242268A (en) * | 2019-09-05 | 2020-06-05 | 中国科学院计算技术研究所 | Method for searching convolutional neural network |
CN115062537A (en) * | 2022-06-06 | 2022-09-16 | 中国人民武装警察部队工程大学 | Photovoltaic cell model parameter identification method based on drunken Chinese strolling chaos wolf group algorithm |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102567781A (en) * | 2011-12-20 | 2012-07-11 | 西北师范大学 | Network behavior prediction method based on neural network |
CN104794526A (en) * | 2015-05-13 | 2015-07-22 | 哈尔滨工程大学 | Automatic ship anti-collision method optimized by wolf colony search algorithm |
US20160070956A1 (en) * | 2014-09-05 | 2016-03-10 | Huawei Technologies Co., Ltd. | Method and Apparatus for Generating Facial Feature Verification Model |
CN106022517A (en) * | 2016-05-17 | 2016-10-12 | 温州大学 | Risk prediction method and device based on nucleus limit learning machine |
CN107085942A (en) * | 2017-06-26 | 2017-08-22 | 广东工业大学 | A kind of traffic flow forecasting method based on wolf pack algorithm, apparatus and system |
CN107832830A (en) * | 2017-11-17 | 2018-03-23 | 湖北工业大学 | Intruding detection system feature selection approach based on modified grey wolf optimized algorithm |
-
2018
- 2018-04-10 CN CN201810316604.5A patent/CN108491923B/en not_active Expired - Fee Related
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102567781A (en) * | 2011-12-20 | 2012-07-11 | 西北师范大学 | Network behavior prediction method based on neural network |
US20160070956A1 (en) * | 2014-09-05 | 2016-03-10 | Huawei Technologies Co., Ltd. | Method and Apparatus for Generating Facial Feature Verification Model |
CN104794526A (en) * | 2015-05-13 | 2015-07-22 | 哈尔滨工程大学 | Automatic ship anti-collision method optimized by wolf colony search algorithm |
CN106022517A (en) * | 2016-05-17 | 2016-10-12 | 温州大学 | Risk prediction method and device based on nucleus limit learning machine |
CN107085942A (en) * | 2017-06-26 | 2017-08-22 | 广东工业大学 | A kind of traffic flow forecasting method based on wolf pack algorithm, apparatus and system |
CN107832830A (en) * | 2017-11-17 | 2018-03-23 | 湖北工业大学 | Intruding detection system feature selection approach based on modified grey wolf optimized algorithm |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111242268A (en) * | 2019-09-05 | 2020-06-05 | 中国科学院计算技术研究所 | Method for searching convolutional neural network |
CN115062537A (en) * | 2022-06-06 | 2022-09-16 | 中国人民武装警察部队工程大学 | Photovoltaic cell model parameter identification method based on drunken Chinese strolling chaos wolf group algorithm |
CN115062537B (en) * | 2022-06-06 | 2024-04-12 | 中国人民武装警察部队工程大学 | Photovoltaic cell model parameter identification method based on drunken walk chaos wolf group algorithm |
Also Published As
Publication number | Publication date |
---|---|
CN108491923B (en) | 2021-09-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Kulkarni | Crop disease detection using deep learning | |
Han et al. | A survey on metaheuristic optimization for random single-hidden layer feedforward neural network | |
Swasono et al. | Classification of tobacco leaf pests using VGG16 transfer learning | |
WO2022121289A1 (en) | Methods and systems for mining minority-class data samples for training neural network | |
CN107818302A (en) | Non-rigid multi-scale object detection method based on convolutional neural network | |
CN108334949A (en) | A kind of tachytelic evolution method of optimization depth convolutional neural networks structure | |
CN112581262A (en) | Whale algorithm-based fraud detection method for optimizing LVQ neural network | |
Shrivastava et al. | Performance evaluation of feed-forward neural network with soft computing techniques for hand written English alphabets | |
Abd-Alsabour | A review on evolutionary feature selection | |
Asadi et al. | ACORI: A novel ACO algorithm for rule induction | |
CN110287985B (en) | Depth neural network image identification method based on variable topology structure with variation particle swarm optimization | |
CN111105045A (en) | Method for constructing prediction model based on improved locust optimization algorithm | |
CN116015967B (en) | Industrial Internet intrusion detection method based on improved whale algorithm optimization DELM | |
Zhang et al. | Evolving neural network classifiers and feature subset using artificial fish swarm | |
Kolli et al. | Plant disease detection using convolutional neural network | |
CN108491923A (en) | Based on the pest image-recognizing method for improving wolf pack algorithm optimization Bayesian network | |
CN109300140A (en) | Method for detecting image edge based on genetic algorithm and ant group algorithm | |
CN115643189A (en) | Network anomaly detection method based on group intelligent algorithm and isolated forest | |
Punitha et al. | A novel breast cancer diagnosis scheme with intelligent feature and parameter selections | |
CN115050022A (en) | Crop pest and disease identification method based on multi-level self-adaptive attention | |
Ashwini et al. | An optimal model for identification and classification of corn leaf disease using hybrid 3D-CNN and LSTM | |
CN116612386A (en) | Pepper disease and pest identification method and system based on hierarchical detection double-task model | |
Vora et al. | An ensemble of convolutional neural networks to detect foliar diseases in apple plants | |
Dominguez-Catena et al. | Additional feature layers from ordered aggregations for deep neural networks | |
Chen et al. | KDT-SPSO: A multimodal particle swarm optimisation algorithm based on kd trees for palm tree detection |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20210924 |
|
CF01 | Termination of patent right due to non-payment of annual fee |