CN110288606A - A kind of three-dimensional grid model dividing method of the extreme learning machine based on ant lion optimization - Google Patents
A kind of three-dimensional grid model dividing method of the extreme learning machine based on ant lion optimization Download PDFInfo
- Publication number
- CN110288606A CN110288606A CN201910576352.4A CN201910576352A CN110288606A CN 110288606 A CN110288606 A CN 110288606A CN 201910576352 A CN201910576352 A CN 201910576352A CN 110288606 A CN110288606 A CN 110288606A
- Authority
- CN
- China
- Prior art keywords
- ant
- grid model
- dimensional grid
- ant lion
- lion
- 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
- 241001206881 Myrmeleon inconspicuus Species 0.000 title claims abstract description 98
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000005457 optimization Methods 0.000 title claims abstract description 36
- 230000011218 segmentation Effects 0.000 claims abstract description 59
- 238000012360 testing method Methods 0.000 claims abstract description 58
- 238000012549 training Methods 0.000 claims abstract description 39
- 239000011159 matrix material Substances 0.000 claims abstract description 38
- 238000000926 separation method Methods 0.000 claims abstract description 20
- 241000238413 Octopus Species 0.000 claims abstract description 16
- 230000004913 activation Effects 0.000 claims description 15
- 238000005295 random walk Methods 0.000 claims description 12
- 241000282320 Panthera leo Species 0.000 claims description 10
- 230000006870 function Effects 0.000 claims description 6
- 238000002372 labelling Methods 0.000 claims description 6
- 210000002569 neuron Anatomy 0.000 claims description 6
- 230000003044 adaptive effect Effects 0.000 claims description 3
- 230000000694 effects Effects 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 238000012800 visualization Methods 0.000 claims description 3
- 230000008901 benefit Effects 0.000 abstract description 4
- 230000008859 change Effects 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 239000012141 concentrate Substances 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 241000251468 Actinopterygii Species 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- 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
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses the three-dimensional grid model dividing methods of the extreme learning machine optimized based on ant lion, compared with prior art, the present invention is achieved the object of the present invention using step 1 to step 7, the advantage is that: using ant in ant lion optimization algorithm by the double influence of elite ant lion and roulette strategy, iteration updates and optimal solution is assigned to elite ant lion, to improve and optimizate the input weight and hidden layer bias matrix that extreme learning machine generates at random, training obtains a high-precision segmentation and classification device and tests Princeton model.Compared to the work of forefathers, on 6 class model data set of Airplane, Ant, Chair, Octopus, Teddy and Fish, model algorithm time-consuming 1000s or so of the training dough sheet number between 20k-30k, and the dividing method than being not optimised obtains the higher accuracy of separation, the highest accuracy of separation reaches 99.49%.
Description
Technical field
The invention belongs to 3 d image data processing technology fields, and in particular to a kind of limit study based on ant lion optimization
The three-dimensional grid model dividing method of machine, this method are used to carry out semantic segmentation to existing three-dimensional grid model.
Background technique
Have benefited from the rapid development of hardware sensor and reconstruction technique, the quantity of miscellaneous digital multimedia data is not
It is disconnected to increase.It is analyzed and is handled for 3 d image data, become the hot spot of computer graphics study.Wherein grid model
Segmentation be research topic that Fundamentals of Computer Graphics is also importance.The segmentation of grid model is the geometry according to model
Or topological characteristic, it is decomposed, becomes certain amount and communicate with each other, each part has simple shape meaning sub-grid
The work of piece.Researchers at home and abroad propose many methods for grid model segmentation.According to whether the label of model is made
To input information, the segmentation of grid model can be divided into: segmentation and unsupervised segmentation two major classes based on supervision.Due to based on prison
The very few manual intervention of the dividing method superintended and directed, shows higher segmentation precision.In recent years, researcher proposed in succession
Dividing method based on supervision.But disappear currently based on the grid model dividing method of the deep learning training segmentation and classification device time
Consumption is especially big, how effectively to reduce the training time, and further increases the accuracy of grid model segmentation, is to need at present
The problem of focusing on solving.For some emerging Swarm Intelligent Algorithms, not only simple and flexible but also part can be effectively avoided
It is optimal, so that it is used widely in solving optimization problem.How by colony intelligence optimization theory be applied to threedimensional model
It is the hot issue for needing further to research and solve in dividing processing.
Summary of the invention
It is king-sized in order to solve the grid model dividing method training segmentation and classification device time loss based on deep learning
Disadvantage, and the accuracy of grid model segmentation is further increased, this paper presents a kind of limit study based on ant lion optimization
The three-dimensional grid model dividing method of machine.
A kind of three-dimensional grid model dividing method of the extreme learning machine based on ant lion optimization, it is described to be optimized based on ant lion
The three-dimensional grid model dividing method of extreme learning machine comprise the steps of:
Step 1, for Princeton model data concentrate Airplane, Ant, Chair, Octopus, Teddy,
6 class three-dimensional grid model of Fish carries out dough sheet feature extraction to each three-dimensional grid model, and is normalized, normalizing
Change between [- 1,1], eliminate dimension impact, wherein dough sheet feature, comprising: dough sheet curvature feature, PCA feature, in shape under
Literary feature, average geodesic distance feature, shape characteristics of diameters.
Step 2, Airplane, Ant, Chair, Octopus, Teddy, Fish Princeton model data concentrated
6 class three-dimensional grid models, are divided into training dataset and test data set in proportion respectively, due in the data set of Princeton
6 class three-dimensional grid model of Airplane, Ant, Chair, Octopus, Teddy, Fish, the quantity of every class grid model are all
20, random selection 95% namely 19 models are as training dataset, for training segmentation and classification device, remaining 5%,
That is 1 model is as test data set, for testing to trained segmentation and classification device.
Step 3, the input feature value that obtained every class three-dimensional grid model training dataset is divided based on step 2 is established
P, specifically: every class three-dimensional grid model training data is concentrated to the model for including, in step 1 the dough sheet after normalized
Feature merges, and forms input feature value P, is expressed as P={ (x for the input feature value P that sample number is Mj,tj),j
=1,2 ..., M }, wherein xj=(xj1,xj2,…,xjn)T∈Rn, indicate to be directed to the extracted dough sheet feature of three-dimensional grid model, n
For the attribute size of input feature value P, tj=(tj1,tj2,…,tjm)T∈RmOr it is expressed as matrixIt represents
The label that three-dimensional grid model is beaten.
Step 4, the testing feature vector that obtained every class three-dimensional grid model test data set is divided based on step 2 is established
P', specifically: the dough sheet feature after the three-dimensional grid model of test data concentration in step 1 normalized is formed into test
Feature vector P'.
Step 5, the size of the hidden neuron L parameter of extreme learning machine model is set, specifically: by hidden neuron
Number L is set as 300, and basic extreme learning machine algorithm is realized by step 5.1 to step 5.3:
Step 5.1, random to generate input weight wiB is biased with hidden layeri;
Step 5.2, by formulaHidden layer output matrix H is calculated, wherein
G () indicates activation primitive;
Step 5.3, by formulaOutput weight matrix is calculated, wherein H-It is H generalized inverse matrix, matrixRepresent the label that three-dimensional grid model is beaten.
Step 6, extreme learning machine is optimized using ant lion optimization algorithm, maximum number of iterations in ant lion optimization algorithm is set
ImaxIt is 20 for 25 and population scale N, using the input feature value P obtained based on step 3 as input data, is input to ant lion
It is trained in the extreme learning machine model of optimization, training obtains optimal segmentation and classification device, and the survey to obtaining based on step 4
Examination feature vector P' is split the specific of test and executes operation, is realized by step 6.1 to step 6.6:
Step 6.1, random initializtion ant and ant lion population, according to formulaIn limitation
The size of condition, random initializtion ant and ant lion population is N × L × (n+1), each ant and ant lion individual in population
Position indicates extreme learning machine parameter combination f (w to be optimizedi,bi);
Step 6.2, the fitness value of every ant lion is calculated, and optimal is considered as elite ant lion, extremely by step 6.2.1
Step 6.2.7 is realized:
Each ant lion individual in ant lion population is divided into the input weight w that size is L × n by step 6.2.1i, size
It is the hidden layer biasing b of L × 1i;
Step 6.2.2, by formulaHidden layer output matrix H is calculated,
Middle g () indicates that activation primitive, the activation primitive that this method uses are S type function;
Step 6.2.3, by formulaOutput weight matrix is calculated, wherein H-It is H generalized inverse matrix, matrixRepresent the label that three-dimensional grid model is beaten;
Step 6.2.4, the segmentation and classification device obtained by input feature value P training, wherein segmentation and classification device includes input
Weight, hidden layer biasing, exports weight, predicts the label of each dough sheet of testing feature vector P' three-dimensional grid model;
The label that prediction obtains is compared with the label of standard, counts the number of correct labeling by step 6.2.5;
Step 6.2.6, calculates the fitness value of ant lion, i.e. the segmentation of testing feature vector P' three-dimensional grid model is accurate
Degree, the accuracy of separation are the ratio between the grid surface number being correctly marked in grid model and total grid surface number;
Step 6.2.7 compares the size of the fitness value of each individual in ant lion population, from big to small according to fitness value
Sequence, is assigned to elite ant lion for optimal;
Step 6.3, t=1 is enabled, into iterative cycles, updates the position f (w of anti,bi), by step 6.3.1 to step
6.3.3 it realizes:
Step 6.3.1 calculates influence of the roulette strategy to ant random walk, obtains roulette strategy and changes at the t times
For when the random walk that selects
Step 6.3.2 calculates influence of the elite ant lion to ant random walk, obtains when the t times iteration ant in elite
Random walk around ant lion
Step 6.3.3, by formulaIt is common by wheel disc roulette wheel strategy and elite ant lion to calculate every ant
Walking about under effect, updates the position f (w of anti,bi);
Step 6.4, the fitness value for calculating every ant is realized by step 6.4.1 to step 6.4.6:
Each ant individual in ant population is divided into the input weight w that size is L × n by step 6.4.1i, size
It is the hidden layer biasing b of L × 1i;
Step 6.4.2, by formulaHidden layer output matrix H is calculated,
Middle g () indicates that activation primitive, the activation primitive that this method uses are S type function;
Step 6.4.3, by formulaOutput weight matrix is calculated, wherein H-It is H generalized inverse matrix;
Step 6.4.4, the segmentation and classification device obtained by input feature value P training, segmentation and classification device include input weight,
Hidden layer biasing, exports weight, predicts the label of each dough sheet of testing feature vector P' three-dimensional grid model;
The label that prediction obtains is compared with the label of standard, counts the number of correct labeling by step 6.4.5;
Step 6.4.6, calculates the fitness value of ant, i.e. the segmentation of testing feature vector P' three-dimensional grid model is accurate
Degree, the accuracy of separation are the ratio between the grid surface number being correctly marked in grid model and total grid surface number;
Step 6.5, elite ant lion is updated, is realized by step 6.5.1 to step 6.5.2:
Step 6.5.1, ant has better fitness value than ant lion, by formulaAnt lion is updated into the position f (w to captured anti,bi), conversely, then ant
The position of lion remains unchanged;
Step 6.5.2, ant lion are more preferable than elite ant lion fitness value, the position of elite ant lion are updated, conversely, then elite ant
The position of lion remains unchanged;
Step 6.6, judge whether ant lion optimization algorithm reaches maximum number of iterations ImaxIf reached, elite ant is just exported
The accuracy of separation and the position of the corresponding adaptive optimal control angle value of lion namely testing feature vector P' three-dimensional grid model, Ye Jiji
Limit learning machine parameter combination f (w to be optimizedi,bi), otherwise the number of iterations t adds 1, jumps back to step 6.3 and continues to execute, until reaching
The maximum number of iterations I of algorithmmax;
Step 7, the optimal segmentation and classification device obtained based on step 6 training, and to the test feature obtained based on step 4
Vector P' is split test, obtains the optimal accuracy of separation of testing feature vector P' three-dimensional grid model, and it is right therewith to generate
The three-dimensional grid model segmentation result .seg file answered, and obtain visualization result.
The beneficial effects of the present invention are:
Compared with prior art, combining step 1 to step 7, remarkable advantage of the invention is: utilizing ant lion optimization algorithm
Middle ant by elite ant lion and roulette strategy double influence, iteration update simultaneously optimal solution is assigned to elite ant lion, thus
The input weight and hidden layer bias matrix that extreme learning machine generates at random are improved and optimizated, training obtains a high-precision segmentation
Classifier tests Princeton model.Compared to the work of forefathers, Airplane, Ant, Chair, Octopus,
On 6 class model data set of Teddy and Fish, model algorithm time-consuming 1000s or so of the training dough sheet number between 20k-30k, and
Dividing method than being not optimised obtains the higher accuracy of separation, and the highest accuracy of separation reaches 99.49%.
Detailed description of the invention
Fig. 1 is the overall flow figure of dividing method of the present invention;
Fig. 2 is basic extreme learning machine algorithm principle figure;
The convergence figure of Fig. 3 Airplane model aircraft different population scale;
The convergence figure of Fig. 4 Octopus octopus model different population scale;
Fig. 5 is Airplane model aircraft segmentation result figure;
Fig. 6 is Ant ant model segmentation result figure;
Fig. 7 is Chair chair model segmentation result figure;
Fig. 8 is Octopus octopus model segmentation result figure;
Fig. 9 is Teddy teddy bear model segmentation result figure;
Figure 10 is Fish fish model segmentation result figure.
Specific implementation method
Present invention is further described in detail with example with reference to the accompanying drawing.
As shown in Figure 1, being that the present invention is based on the streams of the three-dimensional grid model dividing method of the extreme learning machine of ant lion optimization
Cheng Tu, can summary be step 1 to step 7:
Step 1, for Princeton model data concentrate Airplane, Ant, Chair, Octopus, Teddy,
6 class three-dimensional grid model of Fish carries out dough sheet feature extraction to each three-dimensional grid model, and is normalized, normalizing
Change between [- 1,1], eliminate dimension impact, wherein dough sheet feature, comprising: dough sheet curvature feature, PCA feature, in shape under
Literary feature, average geodesic distance feature, shape characteristics of diameters.
Step 2, Airplane, Ant, Chair, Octopus, Teddy, Fish Princeton model data concentrated
6 class three-dimensional grid models, are divided into training dataset and test data set in proportion respectively, due in the data set of Princeton
6 class three-dimensional grid model of Airplane, Ant, Chair, Octopus, Teddy, Fish, the quantity of every class grid model are all
20, random selection 95% namely 19 models are as training dataset, for training segmentation and classification device, remaining 5%,
That is 1 model is as test data set, for testing to trained segmentation and classification device.
Step 3, the input feature value that obtained every class three-dimensional grid model training dataset is divided based on step 2 is established
P, specifically: every class three-dimensional grid model training data is concentrated to the model for including, in step 1 the dough sheet after normalized
Feature merges, and forms input feature value P, is expressed as P={ (x for the input feature value P that sample number is Mj,tj),j
=1,2 ..., M }, wherein xj=(xj1,xj2,…,xjn)T∈Rn, indicate to be directed to the extracted dough sheet feature of three-dimensional grid model, n
For the attribute size of input feature value P, tj=(tj1,tj2,…,tjm)T∈RmOr it is expressed as matrixIt represents
The label that three-dimensional grid model is beaten.
Step 4, the testing feature vector that obtained every class three-dimensional grid model test data set is divided based on step 2 is established
P', specifically: the dough sheet feature after the three-dimensional grid model of test data concentration in step 1 normalized is formed into test
Feature vector P'.
Step 5, the size of the hidden neuron L parameter of extreme learning machine model is set, specifically: by hidden neuron
Number L is set as 300, and basic extreme learning machine algorithm principle by step 5.1 to step 5.3 as shown in Fig. 2, realized:
Step 5.1, random to generate input weight wiB is biased with hidden layeri;
Step 5.2, by formulaHidden layer output matrix H is calculated, wherein
G () indicates activation primitive;
Step 5.3, by formulaOutput weight matrix is calculated, wherein H- is H generalized inverse matrix, matrixRepresent the label that three-dimensional grid model is beaten.
Step 6, extreme learning machine is optimized using ant lion optimization algorithm, first in selection Princeton data set
The model of 2 classifications of Airplane, Octopus carries out test experiments, Population Size N is respectively set to 10,20 and 30, such as
Fig. 3-Fig. 4, experiment have counted the convergence of this 2 kinds of models, 50 iteration different population scale respectively, due to segmentation side of the invention
The runing time of method can double to increase with the increase of population scale, while consider runing time and the accuracy of separation, work as kind
Group scale N is when being set as 20, we at runtime between good balance is achieved between the accuracy of separation, when iterating to the 25th
After secondary or so, accuracy of separation kept stable, so maximum number of iterations I in setting ant lion optimization algorithmmaxFor 25 Hes
Population scale N is 20, using the input feature value P obtained based on step 3 as input data, is input to the limit of ant lion optimization
It is trained in learning machine (ALO-ELM) model, training obtains optimal segmentation and classification device, and to being obtained based on step 4
Testing feature vector P' is split the specific of test and executes operation, is realized by step 6.1 to step 6.6:
Step 6.1, random initializtion ant and ant lion population, according to formulaIn limitation
The size of condition, random initializtion ant and ant lion population is N × L × (n+1), each ant and ant lion individual in population
Position indicates extreme learning machine parameter combination f (w to be optimizedi,bi);
Step 6.2, the fitness value of every ant lion is calculated, and optimal is considered as elite ant lion, extremely by step 6.2.1
Step 6.2.7 is realized:
Each ant lion individual in ant lion population is divided into the input weight w that size is L × n by step 6.2.1i, size
It is the hidden layer biasing b of L × 1i;
Step 6.2.2, by formulaHidden layer output matrix H is calculated,
Middle g () indicates that activation primitive, the activation primitive that this method uses are S type function;
Step 6.2.3, by formulaOutput weight matrix is calculated, wherein H-It is H generalized inverse matrix, matrixRepresent the label that three-dimensional grid model is beaten;
Step 6.2.4, the segmentation and classification device obtained by input feature value P training, wherein segmentation and classification device includes input
Weight, hidden layer biasing, exports weight, predicts the label of each dough sheet of testing feature vector P' three-dimensional grid model;
The label that prediction obtains is compared with the label of standard, counts the number of correct labeling by step 6.2.5;
Step 6.2.6, calculates the fitness value of ant lion, i.e. the segmentation of testing feature vector P' three-dimensional grid model is accurate
Degree, the accuracy of separation are the ratio between the grid surface number being correctly marked in grid model and total grid surface number;
Step 6.2.7 compares the size of the fitness value of each individual in ant lion population, from big to small according to fitness value
Sequence, is assigned to elite ant lion for optimal;
Step 6.3, t=1 is enabled, into iterative cycles, updates the position f (w of anti,bi), by step 6.3.1 to step
6.3.3 it realizes:
Step 6.3.1 calculates influence of the roulette strategy to ant random walk, obtains roulette strategy and changes at the t times
For when the random walk that selects
Step 6.3.2 calculates influence of the elite ant lion to ant random walk, obtains when the t times iteration ant in elite
Random walk around ant lion
Step 6.3.3, by formulaIt is common by wheel disc roulette wheel strategy and elite ant lion to calculate every ant
Walking about under effect, updates the position f (w of anti,bi);
Step 6.4, the fitness value for calculating every ant is realized by step 6.4.1 to step 6.4.6:
Each ant individual in ant population is divided into the input weight w that size is L × n by step 6.4.1i, size
It is the hidden layer biasing b of L × 1i;
Step 6.4.2, by formulaHidden layer output matrix H is calculated,
Middle g () indicates that activation primitive, the activation primitive that this method uses are S type function;
Step 6.4.3, by formulaOutput weight matrix is calculated, wherein H-It is H generalized inverse matrix;
Step 6.4.4, the segmentation and classification device obtained by input feature value P training, segmentation and classification device include input weight,
Hidden layer biasing, exports weight, predicts the label of each dough sheet of testing feature vector P' three-dimensional grid model;
The label that prediction obtains is compared with the label of standard, counts the number of correct labeling by step 6.4.5;
Step 6.4.6, calculates the fitness value of ant, i.e. the segmentation of testing feature vector P' three-dimensional grid model is accurate
Degree, the accuracy of separation are the ratio between the grid surface number being correctly marked in grid model and total grid surface number;
Step 6.5, elite ant lion is updated, is realized by step 6.5.1 to step 6.5.2:
Step 6.5.1, ant has better fitness value than ant lion, by formulaAnt lion is updated into the position f (w to captured anti,bi), conversely, then ant
The position of lion remains unchanged;
Step 6.5.2, ant lion are more preferable than elite ant lion fitness value, the position of elite ant lion are updated, conversely, then elite ant
The position of lion remains unchanged;
Step 6.6, judge whether ant lion optimization algorithm reaches maximum number of iterations ImaxIf reached, elite ant is just exported
The accuracy of separation and the position of the corresponding adaptive optimal control angle value of lion namely testing feature vector P' three-dimensional grid model, Ye Jiji
Limit learning machine parameter combination f (w to be optimizedi,bi), otherwise the number of iterations t adds 1, jumps back to step 6.3 and continues to execute, until reaching
The maximum number of iterations I of algorithmmax;
Step 7, the optimal segmentation and classification device obtained based on step 6 training, and to the test feature obtained based on step 4
Vector P' is split test, obtains the optimal accuracy of separation of testing feature vector P' three-dimensional grid model, and it is right therewith to generate
The three-dimensional grid model segmentation result .seg file answered, and obtain visualization result, such as Fig. 5-Figure 10, be respectively about
The segmentation result of 6 class three-dimensional grid model of Airplane, Ant, Chair, Octopus, Teddy, Fish.It is visual in the present invention
Change result and Chen X, journey used in Golovinskiy A, Funkhouser T.A benchmark for 3D mesh fruit
Sequence is identical.
For the advantage for showing dividing method of the present invention, experiment segmentation result of the invention is learnt with based on the basic limit
Threedimensional model dividing method CGF [2014], grid model dividing method ATG [2015] and base based on deep layer convolutional network
It is compared in the threedimensional model dividing method CEA [2017] of Fast Learning.
The dividing method of the present invention of table 1 is compared with other automatic Segmentation accuracy based on supervision
From table 1 it follows that dividing method of the present invention has very big promotion in contrast to CGF [2014] segmentation precision, than text
CEA [2017] is offered also to improve.In addition to this, although dividing method partial segmentation result of the present invention is relatively lower than ATG
[2015], it but on time performance, needs to spend 1000s or so in the dough sheet number of training 20k-30k.Same dough sheet number, phase
Than the method time-consuming 4h or so in ATG [2015], advantage is shown.In general, dividing method of the present invention is excellent by colony intelligence
The thought ant lion optimization algorithm of change is applied to the segmentation research of grid model, using this basic model of extreme learning machine,
Deep neural network training time too long problem is effectively prevented, and compared with the work of forefathers, the accuracy of separation keeps higher
Level, algorithm or effective and feasible.
Above embodiments are not limited to the technical solution of the embodiment itself, can be incorporated between embodiment new
Embodiment.The above embodiments are merely illustrative of the technical solutions of the present invention and is not intended to limit it, all without departing from the present invention
Any modification of spirit and scope or same replacement, shall fall within the scope of the technical solution of the present invention.
Claims (10)
1. a kind of three-dimensional grid model dividing method of the extreme learning machine based on ant lion optimization, which is characterized in that comprising following
Step:
Step 1, Airplane, Ant, Chair, Octopus, Teddy, Fish6 class concentrated for Princeton model data
Three-dimensional grid model carries out dough sheet feature extraction to each three-dimensional grid model, and is normalized;
Step 2, Airplane, Ant, Chair, Octopus, Teddy, Fish6 class three Princeton model data concentrated
Grid model is tieed up, is divided into training dataset and test data set in proportion respectively;
Step 3, the input feature value P that obtained every class three-dimensional grid model training dataset is divided based on step 2 is established;
Step 4, the testing feature vector P' that obtained every class three-dimensional grid model test data set is divided based on step 2 is established;
Step 5, the size of the hidden neuron L parameter of extreme learning machine model is set;
Step 6, extreme learning machine is optimized using ant lion optimization algorithm, maximum number of iterations I in ant lion optimization algorithm is setmaxWith
The size of population scale N is input to the pole of ant lion optimization using the input feature value P obtained based on step 3 as input data
Be trained in limit learning machine model, training obtains optimal segmentation and classification device, and to the test feature obtained based on step 4 to
Amount P' is split test;
Step 7, the optimal segmentation and classification device obtained based on step 6 training, and to the testing feature vector obtained based on step 4
P' is split test, obtains the optimal accuracy of separation of testing feature vector P' three-dimensional grid model, generates corresponding
Three-dimensional grid model segmentation result .seg file, and obtain visualization result.
2. the three-dimensional grid model dividing method of the extreme learning machine according to claim 1 based on ant lion optimization, special
Sign is: in the step 1, the dough sheet feature, comprising: dough sheet curvature feature, Shape context feature, is averaged at PCA feature
Geodesic distance feature, shape characteristics of diameters, the normalized normalize between [- 1,1], eliminate dimension impact.
3. the three-dimensional grid model dividing method of the extreme learning machine according to claim 1 based on ant lion optimization, special
Sign is: in the step 2, Airplane, Ant that Princeton model data is concentrated, Chair, Octopus, Teddy,
Fish6 class three-dimensional grid model, is divided into training dataset and test data set in proportion respectively, specifically: Princeton number
According to concentration Airplane, Ant, Chair, Octopus, Teddy, Fish6 class three-dimensional grid model, the quantity of every class grid model
It is all 20, random selection 95% namely 19 models are remaining for training segmentation and classification device as training dataset
5% namely 1 model as test data set, for testing to trained segmentation and classification device.
4. the three-dimensional grid model dividing method of the extreme learning machine according to claim 1 based on ant lion optimization, special
Sign is: in the step 3, it is special to establish the input for dividing obtained every class three-dimensional grid model training dataset based on step 2
Vector P is levied, specifically: every class three-dimensional grid model training data is concentrated to the model for including, in step 1 after normalized
Dough sheet feature merge, form input feature value P, for sample number be M input feature value P be expressed as P=
{(xj,tj), j=1,2 ..., M }, wherein xj=(xj1,xj2,…,xjn)T∈Rn, indicate extracted for three-dimensional grid model
Dough sheet feature, n are the attribute size of input feature value P, tj=(tj1,tj2,…,tjm)T∈RmOr it is expressed as matrixRepresent the label that three-dimensional grid model is beaten.
5. the three-dimensional grid model dividing method of the extreme learning machine according to claim 1 based on ant lion optimization, special
Sign is: in the step 4, it is special to establish the test for dividing obtained every class three-dimensional grid model test data set based on step 2
Vector P' is levied, specifically: by the dough sheet feature group after the three-dimensional grid model of test data concentration in step 1 normalized
At testing feature vector P'.
6. the three-dimensional grid model dividing method of the extreme learning machine according to claim 1 based on ant lion optimization, special
Sign is: in the step 5, the size of the hidden neuron L parameter of extreme learning machine model is set, specifically: by hidden layer mind
300 are set as through first number L,;The extreme learning machine rudimentary algorithm is realized by step 5.1 to step 5.3:
Step 5.1, random to generate input weight wiB is biased with hidden layeri;
Step 5.2, by formulaHidden layer output matrix H is calculated, wherein g ()
Indicate activation primitive;
Step 5.3, by formulaOutput weight matrix is calculated, wherein H-It is H generalized inverse matrix, matrix
Represent the label that three-dimensional grid model is beaten.
7. the three-dimensional grid model dividing method of the extreme learning machine according to claim 1 based on ant lion optimization, special
Sign is: in the step 6, optimizing extreme learning machine using ant lion optimization algorithm, greatest iteration in ant lion optimization algorithm is arranged
Number ImaxIt is 20 for 25 and population scale N, using the input feature value P obtained based on step 3 as input data, is input to
It is trained in the extreme learning machine model of ant lion optimization, training obtains optimal segmentation and classification device, and obtains to based on step 4
Testing feature vector P' be split test specifically execute operation, realized by step 6.1 to step 6.6:
Step 6.1, random initializtion ant and ant lion population, according to formulaIn limitation item
The size of part, random initializtion ant and ant lion population is N × L × (n+1), the position of each ant and ant lion individual in population
Setting indicates extreme learning machine parameter combination f (w to be optimizedi,bi);
Step 6.2, the fitness value of every ant lion is calculated, and optimal is considered as elite ant lion;
Step 6.3, t=1 is enabled, into iterative cycles, updates the position f (w of anti,bi);
Step 6.4, the fitness value of every ant is calculated;
Step 6.5, elite ant lion is updated, is realized by step 6.5.1 to step 6.5.2:
Step 6.5.1, ant has better fitness value than ant lion, by formulaAnt lion is updated into the position f (w to captured anti,bi), conversely, then ant
The position of lion remains unchanged;
Step 6.5.2, ant lion be more preferable than elite ant lion fitness value, updates the position of elite ant lion, conversely, then elite ant lion
Position remains unchanged;
Step 6.6, judge whether ant lion optimization algorithm reaches maximum number of iterations ImaxIf reached, elite ant lion pair is just exported
The accuracy of separation of the adaptive optimal control angle value namely testing feature vector P' three-dimensional grid model answered and position namely the limit
Habit machine parameter combination f (w to be optimizedi,bi), otherwise the number of iterations t adds 1, jumps back to step 6.3 and continues to execute, until reaching algorithm
Maximum number of iterations Imax。
8. the three-dimensional grid model dividing method of the extreme learning machine according to claim 7 based on ant lion optimization, special
Sign is: in the step 6.2, calculating the fitness value of every ant lion, and optimal is considered as elite ant lion, specifically by step
6.2.1 it is realized to step 6.2.7:
Each ant lion individual in ant lion population is divided into the input weight w that size is L × n by step 6.2.1i, size be L ×
1 hidden layer biases bi;
Step 6.2.2, by formulaHidden layer output matrix H is calculated, wherein g
() indicates that activation primitive, the activation primitive that this method uses are S type function;
Step 6.2.3, by formulaOutput weight matrix is calculated, wherein H-It is H generalized inverse matrix, matrixRepresent the label that three-dimensional grid model is beaten;
Step 6.2.4, the segmentation and classification device obtained by input feature value P training, wherein segmentation and classification device includes input weight,
Hidden layer biasing, exports weight, predicts the label of each dough sheet of testing feature vector P' three-dimensional grid model;
The label that prediction obtains is compared with the label of standard, counts the number of correct labeling by step 6.2.5;
Step 6.2.6 calculates the fitness value of ant lion, the i.e. accuracy of separation of testing feature vector P' three-dimensional grid model, point
Cutting accuracy is the ratio between the grid surface number being correctly marked in grid model and total grid surface number;
Step 6.2.7, compares the size of the fitness value of each individual in ant lion population, arranges from big to small according to fitness value
Optimal is assigned to elite ant lion by sequence.
9. the three-dimensional grid model dividing method of the extreme learning machine according to claim 7 based on ant lion optimization, special
Sign is: in the step 6.3, enabling t=1, into iterative cycles, updates the position f (w of anti,bi), specifically by step
6.3.1 it is realized to step 6.3.3:
Step 6.3.1 calculates influence of the roulette strategy to ant random walk, obtains roulette strategy in the t times iteration
The random walk of selection
Step 6.3.2 calculates influence of the elite ant lion to ant random walk, obtains when the t times iteration ant in elite ant lion
The random walk of surrounding
Step 6.3.3, by formulaIt calculates every ant and passes through wheel disc roulette wheel strategy and elite ant lion collective effect
Under walk about, update the position f (w of anti,bi)。
10. the three-dimensional grid model dividing method of the extreme learning machine according to claim 7 based on ant lion optimization, special
Sign is: in the step 6.4, the fitness value of every ant calculated, is specifically realized by step 6.4.1 to step 6.4.6:
Each ant individual in ant population is divided into the input weight w that size is L × n by step 6.4.1i, size be L ×
1 hidden layer biases bi;
Step 6.4.2, by formulaHidden layer output matrix H is calculated, wherein g
() indicates that activation primitive, the activation primitive that this method uses are S type function;
Step 6.4.3, by formulaOutput weight matrix is calculated, wherein H- is H generalized inverse matrix;
Step 6.4.4, the segmentation and classification device obtained by input feature value P training, segmentation and classification device include input weight, are implied
Layer biasing, exports weight, predicts the label of each dough sheet of testing feature vector P' three-dimensional grid model;
The label that prediction obtains is compared with the label of standard, counts the number of correct labeling by step 6.4.5;
Step 6.4.6 calculates the fitness value of ant, the i.e. accuracy of separation of testing feature vector P' three-dimensional grid model, point
Cutting accuracy is the ratio between the grid surface number being correctly marked in grid model and total grid surface number.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910576352.4A CN110288606B (en) | 2019-06-28 | 2019-06-28 | Three-dimensional grid model segmentation method of extreme learning machine based on ant lion optimization |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910576352.4A CN110288606B (en) | 2019-06-28 | 2019-06-28 | Three-dimensional grid model segmentation method of extreme learning machine based on ant lion optimization |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110288606A true CN110288606A (en) | 2019-09-27 |
CN110288606B CN110288606B (en) | 2024-04-09 |
Family
ID=68019586
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910576352.4A Active CN110288606B (en) | 2019-06-28 | 2019-06-28 | Three-dimensional grid model segmentation method of extreme learning machine based on ant lion optimization |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110288606B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112307622A (en) * | 2020-10-30 | 2021-02-02 | 中国兵器科学研究院 | Autonomous planning system and planning method for generating military forces by computer |
CN118229720A (en) * | 2024-05-24 | 2024-06-21 | 长春理工大学 | Multi-threshold image segmentation method suitable for detecting coverage rate of protective cultivated straw |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106228184A (en) * | 2016-07-19 | 2016-12-14 | 东北大学 | A kind of based on the blast furnace fault detection system and the method that optimize extreme learning machine |
CN106650916A (en) * | 2016-12-29 | 2017-05-10 | 西安思源学院 | Grid segmentation method based on ant colony optimization |
US20170161606A1 (en) * | 2015-12-06 | 2017-06-08 | Beijing University Of Technology | Clustering method based on iterations of neural networks |
CN107505392A (en) * | 2017-07-24 | 2017-12-22 | 清华大学 | Material analysis method and device based on grain surface contact acceleration tactile data |
US20180240018A1 (en) * | 2016-05-19 | 2018-08-23 | Jiangnan University | Improved extreme learning machine method based on artificial bee colony optimization |
CN108805907A (en) * | 2018-06-05 | 2018-11-13 | 中南大学 | A kind of pedestrian's posture multiple features INTELLIGENT IDENTIFICATION method |
CN109583092A (en) * | 2018-11-30 | 2019-04-05 | 中南大学 | A kind of intelligent machine diagnosis method for system fault of multi-level multi-mode feature extraction |
CN109767036A (en) * | 2018-12-28 | 2019-05-17 | 北京航空航天大学 | Support vector machines failure prediction method based on the optimization of adaptive ant lion |
CN109766988A (en) * | 2018-09-28 | 2019-05-17 | 中国人民解放军空军工程大学 | Target cluster dividing method based on chaos ant lion optimization algorithm |
-
2019
- 2019-06-28 CN CN201910576352.4A patent/CN110288606B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170161606A1 (en) * | 2015-12-06 | 2017-06-08 | Beijing University Of Technology | Clustering method based on iterations of neural networks |
US20180240018A1 (en) * | 2016-05-19 | 2018-08-23 | Jiangnan University | Improved extreme learning machine method based on artificial bee colony optimization |
CN106228184A (en) * | 2016-07-19 | 2016-12-14 | 东北大学 | A kind of based on the blast furnace fault detection system and the method that optimize extreme learning machine |
CN106650916A (en) * | 2016-12-29 | 2017-05-10 | 西安思源学院 | Grid segmentation method based on ant colony optimization |
CN107505392A (en) * | 2017-07-24 | 2017-12-22 | 清华大学 | Material analysis method and device based on grain surface contact acceleration tactile data |
CN108805907A (en) * | 2018-06-05 | 2018-11-13 | 中南大学 | A kind of pedestrian's posture multiple features INTELLIGENT IDENTIFICATION method |
CN109766988A (en) * | 2018-09-28 | 2019-05-17 | 中国人民解放军空军工程大学 | Target cluster dividing method based on chaos ant lion optimization algorithm |
CN109583092A (en) * | 2018-11-30 | 2019-04-05 | 中南大学 | A kind of intelligent machine diagnosis method for system fault of multi-level multi-mode feature extraction |
CN109767036A (en) * | 2018-12-28 | 2019-05-17 | 北京航空航天大学 | Support vector machines failure prediction method based on the optimization of adaptive ant lion |
Non-Patent Citations (5)
Title |
---|
MIRJALILI S: "The Ant Lion Optimizer", 《ADVANCES IN ENGINEERING SOFTWARE》, pages 80 - 98 * |
UTKU KOSE: "An Ant-Lion Optimizer-Trained Artificial Neural Network System for Chaotic Electroencephalogram (EEG) Prediction", 《APPLIED SCIENCES》, pages 1 - 32 * |
张耀楠 等: "一种基于蚁群优化的网格分割方法", 《计算机工程》, vol. 44, no. 2, pages 277 - 281 * |
李红岩: "三维形状分割和标注的快速学习方法", 《计算机工程与应用》, vol. 53, no. 11, pages 211 - 216 * |
毛羽 等: "改进人工蜂群优化的极限学习机", 《传感器与微系统》, vol. 37, no. 4, pages 116 - 120 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112307622A (en) * | 2020-10-30 | 2021-02-02 | 中国兵器科学研究院 | Autonomous planning system and planning method for generating military forces by computer |
CN112307622B (en) * | 2020-10-30 | 2024-05-17 | 中国兵器科学研究院 | Autonomous planning system and planning method for generating force by computer |
CN118229720A (en) * | 2024-05-24 | 2024-06-21 | 长春理工大学 | Multi-threshold image segmentation method suitable for detecting coverage rate of protective cultivated straw |
CN118229720B (en) * | 2024-05-24 | 2024-07-16 | 长春理工大学 | Multi-threshold image segmentation method suitable for detecting coverage rate of protective cultivated straw |
Also Published As
Publication number | Publication date |
---|---|
CN110288606B (en) | 2024-04-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107273490B (en) | Combined wrong question recommendation method based on knowledge graph | |
CN106529574B (en) | Image classification method based on sparse autocoder and support vector machines | |
CN106529569B (en) | Threedimensional model triangular facet feature learning classification method and device based on deep learning | |
Li et al. | A novel chaotic particle swarm optimization based fuzzy clustering algorithm | |
CN109389124B (en) | Receipt categories of information recognition methods | |
CN105868775A (en) | Imbalance sample classification method based on PSO (Particle Swarm Optimization) algorithm | |
CN106355192A (en) | Support vector machine method based on chaos and grey wolf optimization | |
Xiao et al. | Based on grid-search and PSO parameter optimization for Support Vector Machine | |
CN108629367A (en) | A method of clothes Attribute Recognition precision is enhanced based on depth network | |
CN104778327B (en) | Aircraft wing Optimization Design based on artificial neural network | |
CN107992895A (en) | A kind of Boosting support vector machines learning method | |
CN103488662A (en) | Clustering method and system of parallelized self-organizing mapping neural network based on graphic processing unit | |
CN109063719A (en) | A kind of image classification method of co-ordinative construction similitude and category information | |
CN103886330A (en) | Classification method based on semi-supervised SVM ensemble learning | |
CN110245252A (en) | Machine learning model automatic generation method based on genetic algorithm | |
CN106503731A (en) | A kind of based on conditional mutual information and the unsupervised feature selection approach of K means | |
CN102842043B (en) | Particle swarm classifying method based on automatic clustering | |
Gallagher | Multi-layer perceptron error surfaces: visualization, structure and modelling | |
CN110288606A (en) | A kind of three-dimensional grid model dividing method of the extreme learning machine based on ant lion optimization | |
CN103971415A (en) | Online marking method for three-dimensional model component | |
CN106529604B (en) | A kind of adaptive image tag Robust Prediction method and system | |
WO2023134142A1 (en) | Multi-scale point cloud classification method and system | |
Pan et al. | A Novel Combinational Convolutional Neural Network for Automatic Food-Ingredient Classification. | |
CN111310811B (en) | Large-scene three-dimensional point cloud classification method based on multi-dimensional feature optimal combination | |
CN106339354B (en) | Based on Methods of High-dimensional Data Visualization in the system for cloud computing for improving PCA |
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 |