CN108491923B - Pest image recognition method based on improved wolf colony algorithm optimized Bayesian network - Google Patents
Pest image recognition method based on improved wolf colony algorithm optimized Bayesian network Download PDFInfo
- Publication number
- CN108491923B CN108491923B CN201810316604.5A CN201810316604A CN108491923B CN 108491923 B CN108491923 B CN 108491923B CN 201810316604 A CN201810316604 A CN 201810316604A CN 108491923 B CN108491923 B CN 108491923B
- Authority
- CN
- China
- Prior art keywords
- wolf
- algorithm
- bayesian network
- bayesian
- improved
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
- 241000282461 Canis lupus Species 0.000 title claims abstract description 133
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 54
- 241000607479 Yersinia pestis Species 0.000 title claims abstract description 52
- 238000000034 method Methods 0.000 title claims abstract description 34
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 22
- 238000012360 testing method Methods 0.000 claims abstract description 16
- 230000006399 behavior Effects 0.000 claims abstract description 15
- 238000012549 training Methods 0.000 claims abstract description 12
- 238000000605 extraction Methods 0.000 claims abstract description 9
- 238000005457 optimization Methods 0.000 claims abstract description 9
- 230000003993 interaction Effects 0.000 claims abstract description 8
- 230000035772 mutation Effects 0.000 claims abstract description 7
- 230000008569 process Effects 0.000 claims abstract description 7
- 230000000739 chaotic effect Effects 0.000 claims abstract description 5
- 238000001514 detection method Methods 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 12
- 238000007476 Maximum Likelihood Methods 0.000 claims description 9
- 238000010845 search algorithm Methods 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 6
- 230000009471 action Effects 0.000 claims description 5
- 238000013507 mapping Methods 0.000 claims description 5
- 238000013459 approach Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 3
- 238000007635 classification algorithm Methods 0.000 claims description 2
- 238000007429 general method Methods 0.000 claims 1
- 238000010276 construction Methods 0.000 abstract description 3
- 230000000694 effects Effects 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 7
- 238000011161 development Methods 0.000 description 5
- 239000000523 sample Substances 0.000 description 3
- 241000238631 Hexapoda Species 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 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
- 230000035943 smell Effects 0.000 description 2
- 241000282421 Canidae Species 0.000 description 1
- 241001481710 Cerambycidae Species 0.000 description 1
- 241000254173 Coleoptera Species 0.000 description 1
- 241000238814 Orthoptera Species 0.000 description 1
- 241000270666 Testudines Species 0.000 description 1
- 238000012271 agricultural production Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 229940079593 drug Drugs 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 238000009940 knitting Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 239000002689 soil Substances 0.000 description 1
- 238000003045 statistical classification method Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 230000004083 survival effect Effects 0.000 description 1
- 208000024891 symptom Diseases 0.000 description 1
Images
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 invention provides a pest image recognition method based on an improved wolf colony algorithm optimized Bayesian network, which comprises the following steps: the method comprises the following steps: an Improved binary Wolf pack Algorithm (IBWCA) is provided, the Wolf pack Algorithm is Improved, a mutation operator is added in a walking behavior of a Wolf, an approximation operator is added in a calling behavior, an interaction operator is added in a surrounding behavior, and a Wolf pack is updated in a chaotic motion mode in the updating process of the Wolf pack; step two: a Bayesian structure learning optimization algorithm (Bayesian Network Construction optimization IBWCA, BNC-IBWCA) based on an improved binary wolf pack algorithm; step three: and (4) combining the convolutional neural network and the Bayesian network to identify and process the pest image. And (4) carrying out feature extraction on the pictures of the training set and the test set by using a pre-trained convolutional neural network, and carrying out classification and identification by using a Bayesian classifier.
Description
Technical Field
The invention relates to a convolutional neural network, a Bayesian network and a wolf pack algorithm.
Background
The problems of crop diseases and insect pests are closely related to the life of people, and the development of agriculture is directly influenced. The most common method for prevention and control of pests is currently the use of pesticides. However, due to the variety of pests, people blindly use pesticides without taking medicines according to symptoms under the condition of lacking identification, so that the pests cannot be effectively treated, soil and environment are often damaged, sustainable development of agricultural production is affected, and the output of crops tends to be negatively increased. The form of pest is various, and traditional pest discernment only relies on the characteristics of naked eye observation pest to discern, has certain subjectivity and limitation to working strength is very big, wastes time and energy. Therefore, how to quickly and efficiently identify pests becomes a crucial step for controlling pests.
The pest identification technology is an important link influencing the rapid development of agriculture and is also a key factor for realizing the modernization of agriculture, and because the defects of artificial pest identification are difficult to meet the high requirements of people on pest identification, a plurality of scholars have already provided a plurality of novel pest identification methods by using various advanced technologies, such as voice identification, environment identification and image identification. The most effective method is to identify by the image of the pest, and the pest identification technology based on the image is a relatively new research field at present, and many experts and scholars at home and abroad are still researching. Current pest identification technologies have certain limitations: (1) the traditional pest identification technology is based on point characteristics and line characteristics of images, and has poor identification effect under the condition of complex illumination conditions or variable photographing angles. (2) The traditional image recognition method only extracts representative features in the input image and has certain limitation.
At present, the machine vision technology has great development prospect in the pest identification technology. The Bayesian network is used as one item of machine learning, the structure of the problem can be directly displayed by using the language of the graph theory, and can be analyzed and utilized according to the principle of the probability theory, so that the complexity of reasoning is reduced. Therefore, the Bayesian classifier has a relatively stable classification effect compared with other classification algorithms in the classification effect. However, the structure learning of the bayesian network has been a problem that is difficult in NP. Here we use a search and scoring based approach. An improved wolf pack algorithm is provided, an updating strategy of a wolf pack is designed to improve the problem of local extreme values, the learning algorithm precision is improved, the improved wolf pack algorithm is suitable for structural learning of a Bayesian network, and the learning process of the Bayesian network structure is converted into the problem of finding the optimal wolf pack. And then the pest is identified by combining the unique advantages of the convolutional neural network in the aspect of feature extraction.
In summary, a pest image recognition method based on an improved wolf colony algorithm optimized Bayesian network is provided, and firstly, a saliency region detection method based on global contrast is used for positioning a pest target, then, a GrabCut algorithm is used for automatically segmenting the region of the pest target, and the region is stored in a data file to form a data set for training and testing a model. And secondly, extracting image characteristics on the training set and the test set by using a pretrained convolutional neural network, and inputting the image characteristics into a Bayesian network. Then, the traditional wolf colony algorithm is improved and used as a search algorithm, and a Bayesian Information Criterion (BIC) is used as a scoring function to learn the structure of the Bayesian network. And then, learning parameters of the Bayesian network by using a Maximum Likelihood (ML) algorithm to form a Bayesian classifier.
The invention content is as follows:
in order to solve the problem of agricultural diseases and insect pests, a pest image identification method based on an improved wolf colony algorithm optimized Bayesian network is provided. The invention mainly comprises the following steps: an improved binary wolf pack algorithm is proposed. And (3) taking the improved binary wolf colony algorithm as a search algorithm to carry out structure learning of the Bayesian network to form a Bayesian classifier. And classifying and identifying the pest images by combining the feature extraction function of the convolutional neural network and the classification function of the Bayesian classifier.
A pest image recognition method based on an improved wolf colony algorithm optimized Bayesian network is characterized by comprising the following steps: at least comprises the following steps:
the method comprises the following steps: an Improved Binary Wolf Colony Algorithm (IBWCA)
The wolf colony algorithm is improved, mutation operators are added in the walking behavior of the wolf, approximation operators are added in the calling behavior, and interaction operators are added in the attacking behavior. And in the updating step of the wolf pack, a new artificial wolf is generated by utilizing a chaotic mapping mode to replace the eliminated artificial wolf.
Step two: bayesian structure learning optimization algorithm (Bayesian Network Construction optimization IBWCA, BNC-IBWCA) based on improved binary wolf pack algorithm
The Bayesian network structure can be represented by a adjacency matrix with its corresponding structure matrix encoded as { x }11,,x12,...,x1n,x21,,x22,...,x2n,...,xn1,,xn2,...,xnn}. And then, an improved binary wolf colony algorithm is used as a search algorithm, BIC is used as a scoring function, and an optimal Bayesian network structure is sought.
Step three: method for identifying and processing pest images by combining convolutional neural network and Bayesian network
The method comprises the steps of extracting features of pictures of a training set and a test set by using a pre-trained convolutional neural network, inputting the extracted feature attributes and classification of the training set, learning a Bayesian network structure by using BNC-IBWCA, and then learning Bayesian network parameters by using ML to form a Bayesian network which is most matched with an input data set and serve as a Bayesian classifier. And inputting the extracted characteristic attributes and the classification on the test set into a Bayesian classifier, and testing the Bayesian classifier.
Has the advantages that:
compared with the prior art, the design scheme of the invention can achieve the following technical effects:
1. compared with the traditional wolf colony algorithm, the improved binary wolf colony algorithm is more suitable for learning a Bayesian network structure, in the wolf detection walking process, the added mutation operator ensures the randomness of wolf detection walking, effectively avoids the algorithm from falling into the local optimal solution, in the calling action, the added approximation operator effectively ensures that the general development changes towards the direction of the global optimal solution, and the added interaction operator in the attack action effectively ensures the information interaction between wolf colony individuals. In the updating step of the wolf group, a new artificial wolf is generated by using a chaotic mapping mode, so that the diversity of the group is ensured.
2. Compared with the traditional Bayesian network structure learning method, the Bayesian network structure learning method based on the wolf colony algorithm utilizes the improved binary wolf colony algorithm as a search algorithm and uses the BIC score as a scoring function to perform Bayesian network structure learning, and converts the Bayesian network structure learning process into a corresponding scoring optimization problem of an adjacent matrix. The Bayesian structure learning method is optimized based on the improved wolf colony algorithm, the search precision and the convergence speed are effectively controlled, the trapping into the local optimal solution can be avoided, the calculation robustness is good, and the accuracy of Bayesian network structure learning is ensured.
3. The pest image recognition of the complex background is carried out by combining the saliency detection, the convolutional neural network and the Bayesian network, the pest target positioning is carried out on the pest in the picture by the saliency region detection method based on the global contrast, then the GrabCT algorithm is adopted to automatically segment the region of the pest target, the processed image is used as an original data set, the problem of the complex background of the pest image can be effectively solved, and the recognition rate is further improved. And inputting the image data into a pre-trained convolutional neural network for feature extraction, taking the extracted feature vector as input, and training and testing the Bayesian classifier. The convolutional neural network does not need complex preprocessing work in the early stage, the images can be directly input, the classification work is completely carried out by a full connection layer, the classification effect is poor, and compared with other classifiers, the Bayesian classifier is the most basic statistical classification method in the design method, the classification effect is accurate, but the extraction of the features is only carried out manually, and the extraction is very inaccurate. Therefore, the invention draws the respective advantages of the convolutional neural network and the Bayesian network, avoids the disadvantages and greatly improves the classification capability.
Description of the drawings:
FIG. 1 is a structural framework diagram of a classification model based on a Bayesian network and a convolutional neural network
FIG. 2 is a flow chart of an improved wolf pack algorithm
FIG. 3 is a flow chart of Bayesian classifier learning
The specific implementation mode is as follows:
the method comprises the following steps: an Improved Binary Wolf Colony Algorithm (IBWCA)
Assuming that the search space is m X m dimension, N represents the number of artificial wolfs in wolf cluster, and the position of the ith artificial wolf is defined as Xi={X11,...,X1m,X21,...X2m,...,Xm1,...,Xmm}. Each artificial wolf represents a feasible solution, and the magnitude of the concentration Y of food he smells represents the degree of goodness of the solution.
Step 1: and (6) initializing a numerical value.
(1) Initializing the number N and location X of wolf clustersi
If the position X of the wolf pack is completely randomly initializediAnd the number N will increase the number of searches for the algorithm, so here we use the mutual information theory to initialize the artificial wolf location in wolf pack, although the mutual information theory could judge the father-son relationship, but it couldJudging whether the 2 variables have dependency relationship, finding out the points with dependency relationship, wherein there may be edges between them, then generating N wolfs and their position X according to the dependency relationshipi。
(2) Maximum number of iterations Kmax
(3) Maximum number of wandering times Tmax
(4) Number of worst wolfs Z
(5) Step of walking stepaStep of the running stepbStep of attackc(step size is integer, representing search precision)
(6) Search direction h
Step 2: production and wandering behavior of wolf
Definition 1: mutation operator Θ ═ (X)i,random(stapa)): wherein, XiPosition code X representing wolf's ii={Xi1,...Xim 2},random(stapa) Indicating random selection of a sta in the position code of a sounding wolfaAt each position, the code is mutated (i.e., changed to 0 if the code is 1 and to 1 if the code is 0).
Selecting the best artificial wolf as the head wolf, all the artificial wolfs except the head wolf as the exploring wolf, and executing the wandering action if Y isi>YleadIf yes, the exploring wolf i replaces the wolf and initiates a calling behavior; if Y isi<=YleadThe wolf probe advances one step in h directions with step lengthaNamely, h times of mutation operator theta is executed on the wolf detection, and the food concentration Y which advances to each direction for one step is recordedNewSelecting the maximum value YNew maxIf Y isNew max>YiThen use YNew maxCorresponding to position XNewReplacing original position XiRepeating the above walking action until Yi>YleadOr the number of wandering times reaches Tmax。
Step 3, calling behavior
Define 2: approximation operator Ψ ═ (X)head,Xm,random(stepb)): wherein XheadPosition knitting for indicating head wolfCode, XmPosition code, random (step) representing the wolf of terryb) Means that a continuous length of step is randomly selected from the wolfbReplaces the code at the same location in the wolf.
The head wolf calls the head wolf to the position of the head wolf continuously approaches, wherein, all wolfs except the head wolf are the head wolf. The wolf of terry in larger step size stepbApproaching a wolf head, i.e. having a wolf head perform the approximation operator Ψ, e.g. Xhead(100101100),Xm(100010101) selecting the 2-6 th position of wolf head to replace the 2-6 th position of wolf head to obtain Xm new(100101101), the step size is 5, and the variation represents the influence and guidance of the global optimal solution on the individual. During the course of the wolf-head approach, if Ym>YheadIf yes, the wolf is updated to be a wolf and a new round of calling behavior is initiated; if Y ism<=YheadThe wolf i continues approaching toward the wolf until the distance between the wolf and the wolf is less than the determination distance dnear。
And 4, step 4: attack behavior
Definition 3: the interaction operator Δ ═ Xy,Xz,random(stepc)): wherein XyAnd XzTwo artificial wolves, random (step) within the scope of attack are shownc) Means that the continuous length of an artificial wolf is randomly selected as stepcThe position code of (2) is exchanged with the code of the same position of another artificial wolf.
Under the command of the wolf, the wolf of terrible and the wolf of spy attack the prey, here we assume that the position X of the wolf of terrible isheadI.e. position X of preyfoodThe distance between the artificial wolfs participating in the attack is in a small area, the distance between the artificial wolfs is very close, and in order to capture food as soon as possible, mutual information sharing is needed, namely, an interaction operator delta is executed, the prey concentration Y is calculated, and if Y is replaced, the mutual information sharing is carried outnewGreater than Y before replacementoldThe previous position code is replaced by the replaced position code, otherwise, the position code is unchanged.
And 5: renewing wolf group
Because of the survival mechanism of 'superior-inferior', food is preferentially distributed to strong wolfs, so that the weak wolfs can be starved, wolf groups are updated, poor Z artificial wolfs are eliminated, and the Z artificial wolfs are randomly generated, and the updating mechanism can keep the diversity of the population and effectively avoid the limitation to partial optimal solution.
The chaos is the intrinsic randomness of a decisive system, all patents are traversed in a specified range according to the self 'rule' without repetition, and the chaos randomness is not reduced along with the increase of the information quantity compared with the common randomness, so that new Z wolfs are generated to replace eliminated weak wolfs in a chaos mapping mode, repeated individuals do not appear in a newly generated population, and the diversity of the population is further improved.
Step 6: judging whether to finish
When the obtained result meets the requirement or reaches the maximum iteration number KmaxThen, the algorithm is ended and the position of the wolf head is output. Otherwise, jump to step 2.
Step two: bayesian structure learning optimization algorithm (Bayesian Network Construction optimization IBWCA, BNC-IBWCA) based on improved wolf colony algorithm
1. Coding mode for learning Bayesian network structure
In a bayesian network structure with n nodes, the bayesian network structure may use an adjacency matrix X ═ XijDenotes whereinIts corresponding structure matrix is coded as { x11,,x12,...,x1n,x21,,x22,...,x2n,...,xn1,,xn2,...,xnn}. I.e. encoding for a bayesian network structure.
2. Scoring function
Here we select a bayesian information criterion, called BIC scoring for short, which is an approximation to the edge likelihood function on the premise of a large sample.
Let G be a variable consisting of n X ═ X1,X2,......XnA Bayesian network structure of each variable XiIs r isiEach value {1,2i},XiFather node Xpa(i) Can take the value of qiThe goodness between the Bayesian network G and the data set Q can be measured using a BIC scoring function.
The first term of the formula represents the degree of fit of the structure to the data, and the second term is a complexity penalty term for the model, avoiding overfitting.
The larger the score function value, the better the performance.
3. Search algorithm
IBWCA is used herein as a search algorithm for learning the bayesian network structure.
Assuming that the search space is m X m dimension, N represents the number of wolfs in wolf group, m represents the number of nodes in Bayesian network structure, and the position of the ith artificial wolf is defined as Xi={X11,...,X1m,X21,...X2m,...,Xm1,...,Xmm}. Each artificial wolf represents a feasible Bayesian network, the food concentration Y smells by the artificial wolf represents the fitting degree of the Bayesian network structure and the test data, and the position of the head wolf after the algorithm is finished is the best Bayesian network structure. The Bayesian network structure diagram corresponding to the position of the artificial wolf is a directed acyclic graph. In the process of Bayesian network structure learning, edges are added, deleted and changed continuously, which may cause a loop to appear, and an invalid solution is generated in an iterative process, thereby destroying the robustness of coding. Therefore, each time the position code of the wolf pack is changed, whether a ring exists in the bayesian network structure corresponding to the changed artificial wolf is judged, whether a ring exists is judged by using a depth-first algorithm (DFS) in the text, if so, the position is abandoned, and the operation is carried out again.
Step three: method for identifying and processing pest images by combining convolutional neural network and Bayesian network
Step 1: feature extraction of pictures in training set and test set by using pre-trained convolutional neural network
The method aims at 40 common pests such as beetles, turtles, locusts, longicorn, moths and the like to be classified and identified, original data are obtained by two modes of manual shooting and collection and search through a search engine, then original images are screened, 9000 pictures are selected, the same pests and different growth stages are treated by simply marking two subclasses of adults and larvae of the same pests to be treated (the pests are simply divided into two classes), then a significance region detection method based on global contrast is used for positioning pest targets, then regions of the pest targets are automatically divided by a GrabCut algorithm, and the divided pests are stored in a data file to form a data set for training and testing a model.
Selecting a convolutional neural network pre-trained on a large data set (pest data set), directly inputting two-dimensional pictures in the data set into a convolutional neural network model by taking as input, and then alternately performing the operations of a plurality of volumes of base layers and pooling layers to extract the characteristics of the images in different aspects.
Step 2: inputting the extracted characteristic attributes and training samples on the training set into a learning Bayes classifier
(1) Learning of Bayesian network structures
Learning of Bayesian network structures with BNC-IBWCA as set forth above
(2) Learning of Bayesian network parameters
The Bayesian network parameter learning is not the research focus of the patent, and the parameter learning is simply carried out by using an ML algorithm.
Maximum Likelihood (ML) estimation is based on conventional statistical analysis, which evaluates how well a sample fits to a model based on likelihood, which is shown in equation 1.
The maximum likelihood estimate is exactly Θ, which is the maximum likelihood function.
Θ=argmaxΘL (Θ, D) formula 2
Step four: and (4) inputting the extracted characteristic attributes and the samples on the test set into a Bayes classifier in a classified manner, and testing the classifier.
Inputting the test data set into a well-learned Bayes classifier, predicting the classification, checking the classification effect of the classifier, and if deviation exists, continuing improvement.
Claims (3)
1. A pest image recognition method based on an improved wolf colony algorithm optimized Bayesian network is characterized by comprising the following steps: at least comprises the following steps:
the method comprises the following steps: on the basis of an original wolf pack algorithm, a new operator and operation are added, a search space is adjusted to be binary coding, a new binary wolf pack algorithm IBWCA is provided, the IBWCA improves the original wolf pack algorithm, a mutation operator is added at a randomly selected position in a position code in a wolf detection walking behavior, an approaching operator approaches to a wolf head in a calling behavior, the approaching operator is composed of a wolf head position code, a wolf head position code and a random length position, an interaction operator is added in a attacking behavior of the wolf head and the wolf detection so that the wolf head and the wolf detection can share position code information, and the new wolf is generated by chaotic mapping in the updating step of the wolf pack to improve population diversity to replace the rejected wolf head;
step two: on the basis of the work of the first step, a Bayesian network structure learning optimization general method is combined, a new Bayesian network structure learning optimization algorithm BNC-IBWCA is provided, wherein an adjacency matrix is adopted for Bayesian network structure coding, a scoring function based on a Bayesian information criterion is designed, the first item of the scoring function is the fitting degree of the structure and data, the second item is a model complexity punishment item, a binary wolf pack algorithm IBWCA is used as a search algorithm, the artificial wolf position corresponds to a feasible Bayesian network structure, and a depth-first algorithm is adopted to judge and discard the Bayesian network structure invalid due to the presence of a ring;
step three: method for identifying and processing pest images by combining convolutional neural network and Bayesian network
The method comprises the steps of extracting features of pictures of a training set and a test set by using a pre-trained convolutional neural network, inputting the extracted feature attributes and classification of the training set, learning the structure of the Bayesian network by using BNC-IBWCA, then learning the parameters of the Bayesian network by using a maximum likelihood algorithm to form the Bayesian network which is most matched with an input data set, using the Bayesian network as a Bayesian classifier, inputting the extracted feature attributes and classification of the test set into the Bayesian classifier, and testing the Bayesian classifier.
2. The pest image recognition method based on improved wolf colony algorithm optimized Bayesian network as claimed in claim 1, wherein: the improved binary wolf pack algorithm provided in the step one is suitable for learning of a Bayesian network structure, and a mutation operator is provided in the walking process of the wolf exploration; in the calling behavior, an approximation operator is proposed, and each wolf is close to the optimal wolf at a certain speed; the interaction operator proposed in the enclosing action enables the wolf pack individuals to gradually move to a better position; in the updating step of the wolf pack, a new wolf is generated by utilizing a chaotic mapping mode to replace the eliminated wolf.
3. The pest image recognition method based on improved wolf colony algorithm optimized Bayesian network as claimed in claim 1, wherein: combining significance detection, carrying out pest image recognition of a complex background by a convolutional neural network and a Bayesian network, using the convolutional neural network only for feature extraction, using the extracted feature vector as input, carrying out classification by using a Bayesian classification algorithm, using a pre-trained convolutional neural network for feature extraction, using the Bayesian network obtained by optimizing Bayesian structure learning by using an improved wolf colony algorithm as a classifier, positioning a pest target by using a significance region detection method based on global contrast, and then automatically segmenting the region of the pest target by using a GrabCut algorithm.
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 CN108491923A (en) | 2018-09-04 |
CN108491923B true 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) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111340179A (en) * | 2019-09-05 | 2020-06-26 | 中国科学院计算技术研究所 | Convolutional neural network topology method |
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 |
Family Cites Families (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 |
CN104239858B (en) * | 2014-09-05 | 2017-06-09 | 华为技术有限公司 | A kind of method and apparatus of face characteristic checking |
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 |
CN107085942B (en) * | 2017-06-26 | 2021-01-26 | 广东工业大学 | Traffic flow prediction method, device and system based on wolf colony algorithm |
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
Also Published As
Publication number | Publication date |
---|---|
CN108491923A (en) | 2018-09-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Kulkarni | Crop disease detection using deep learning | |
Jin et al. | Weed identification using deep learning and image processing in vegetable plantation | |
Chiang et al. | Deep learning-based automated forest health diagnosis from aerial images | |
CN107818302A (en) | Non-rigid multiple dimensioned object detecting method based on convolutional neural networks | |
Khan et al. | Deep learning for apple diseases: classification and identification | |
CN104680193B (en) | Online objective classification method and system based on quick similitude network integration algorithm | |
Pandey et al. | A literature survey of optimization techniques for satellite image segmentation | |
CN108491923B (en) | Pest image recognition method based on improved wolf colony algorithm optimized Bayesian network | |
Kolli et al. | Plant disease detection using convolutional neural network | |
Prasath et al. | IoT-based pest detection and classification using deep features with enhanced deep learning strategies | |
Martins et al. | Deep learning applied to the identification of fruit fly in intelligent traps | |
CN109558803B (en) | SAR target identification method based on convolutional neural network and NP criterion | |
Guerrout et al. | Hidden Markov random fields and swarm particles: a winning combination in image segmentation | |
CN113128653A (en) | Mixed group intelligent deep learning model hyper-parameter optimization method | |
Venkateswaran et al. | Performance analysis of GA and PSO based feature selection techniques for improving classification accuracy in remote sensing images | |
Ashwini et al. | An optimal model for identification and classification of corn leaf disease using hybrid 3D-CNN and LSTM | |
Senthilkumar et al. | Prediction of rice disease using modified feature weighted fuzzy clustering (MFWFC) based segmentation and hybrid classification model | |
Chen et al. | KDT-SPSO: A multimodal particle swarm optimisation algorithm based on kd trees for palm tree detection | |
Dominguez-Catena et al. | Additional feature layers from ordered aggregations for deep neural networks | |
Singh et al. | Tomato crop disease classification using convolution neural network and transfer learning | |
Ponce et al. | Towards the distributed wound treatment optimization method for training CNN models: analysis on the MNIST dataset | |
Suresh et al. | Design an efficient disease monitoring system for paddy leaves based on big data mining | |
Kutlu et al. | An Intelligent Software for Measurements of Biological Materials: BioMorph | |
Vrbančič et al. | Profiling the Arion rufus snails with computer vision | |
Michael et al. | Evaluation of Deep Learning CNN Models with 24 Metrics Using Soybean Crop and Broad-Leaf Weed Classification |
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 |