CN110288606B - Three-dimensional grid model segmentation method of extreme learning machine based on ant lion optimization - Google Patents
Three-dimensional grid model segmentation method of extreme learning machine based on ant lion optimization Download PDFInfo
- Publication number
- CN110288606B CN110288606B CN201910576352.4A CN201910576352A CN110288606B CN 110288606 B CN110288606 B CN 110288606B CN 201910576352 A CN201910576352 A CN 201910576352A CN 110288606 B CN110288606 B CN 110288606B
- Authority
- CN
- China
- Prior art keywords
- ant
- dimensional grid
- grid model
- segmentation
- 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.)
- Active
Links
- 230000011218 segmentation Effects 0.000 title claims abstract description 128
- 241001206881 Myrmeleon inconspicuus Species 0.000 title claims abstract description 95
- 238000000034 method Methods 0.000 title claims abstract description 62
- 238000005457 optimization Methods 0.000 title claims abstract description 36
- 238000012360 testing method Methods 0.000 claims abstract description 61
- 239000011159 matrix material Substances 0.000 claims abstract description 40
- 238000012549 training Methods 0.000 claims abstract description 36
- 241000238413 Octopus Species 0.000 claims abstract description 17
- 230000006870 function Effects 0.000 claims description 21
- 230000004913 activation Effects 0.000 claims description 15
- 238000005295 random walk Methods 0.000 claims description 13
- 241000282320 Panthera leo Species 0.000 claims description 12
- 241000251468 Actinopterygii Species 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 9
- 241000257303 Hymenoptera Species 0.000 claims description 8
- 210000002569 neuron Anatomy 0.000 claims description 7
- 238000010606 normalization Methods 0.000 claims description 6
- 230000009471 action Effects 0.000 claims description 3
- 230000006978 adaptation Effects 0.000 claims description 3
- 230000000007 visual effect Effects 0.000 claims description 3
- 241000258923 Neuroptera Species 0.000 claims 1
- 230000008901 benefit Effects 0.000 abstract description 4
- 230000009977 dual effect Effects 0.000 abstract description 2
- 238000002372 labelling Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 241000894007 species Species 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012800 visualization Methods 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
Compared with the prior art, the invention adopts the steps 1 to 7 to realize the aim of the invention, and has the advantages that: by utilizing the dual influence of elite and ant lion and roulette strategies in an ant lion optimization algorithm, iterative updating is carried out, and an optimal solution is assigned to the elite and ant lion, so that the input weight and hidden layer bias matrix randomly generated by an optimization extreme learning machine are improved, and a high-precision segmentation classifier is trained to test a Prins ston model. Compared with the work of the former, on Airplane, ant, chair, octopus, teddy and Fish6 model data sets, the model algorithm with the training of the number of patches between 20k and 30k takes about 1000 seconds, and the segmentation accuracy is higher than that of an unoptimized segmentation method, and the highest segmentation accuracy reaches 99.49%.
Description
Technical Field
The invention belongs to the technical field of three-dimensional image data processing, and particularly relates to a three-dimensional grid model segmentation method of an extreme learning machine based on ant lion optimization.
Background
Thanks to the rapid development of hardware sensors and reconstruction techniques, the amount of a wide variety of digital multimedia data is increasing. Analysis and processing of three-dimensional image data is becoming a hotspot for computer graphics research. The segmentation of the grid model is an important research topic in computer graphics. The division of the grid model is to decompose the grid model according to the geometric or topological characteristics of the model, and the grid model becomes a certain number and is communicated with each other, and each part has the operation of a simple shape meaning sub-grid sheet. Researchers at home and abroad propose a plurality of methods for dividing a grid model. The segmentation of the mesh model can be divided into, depending on whether the label of the model is taken as input information: two major classes of supervised and unsupervised based segmentation. The segmentation method based on supervision shows higher segmentation precision due to too little manual intervention. In recent years, researchers have successively proposed some supervised-based segmentation methods. However, the current grid model segmentation method based on deep learning has extremely high time consumption for training the segmentation classifier, so that the training time is effectively reduced, the accuracy of grid model segmentation is further improved, and the method is a problem which needs to be solved in an important way at present. For some emerging intelligent optimization algorithms of the group, the method is simple and flexible, and can effectively avoid local optimization, so that the method is widely applied to solving the optimization problem. How to apply the theory of intelligent optimization of groups to the segmentation processing of a three-dimensional model is a hot spot problem which needs to be further researched and solved.
Disclosure of Invention
In order to solve the defect that the time consumption for training a segmentation classifier is extremely large in a mesh model segmentation method based on deep learning and further improve the accuracy of mesh model segmentation, a three-dimensional mesh model segmentation method of an extreme learning machine based on ant lion optimization is provided.
A three-dimensional grid model segmentation method of an extreme learning machine based on ant lion optimization comprises the following steps:
step 1, for Airplane, ant, chair, octopus, teddy, fish three-dimensional grid models in a Prlington model dataset, extracting surface patch characteristics of each three-dimensional grid model, carrying out normalization processing, normalizing to be between [ -1,1] and eliminating dimension influence, wherein the surface patch characteristics comprise: patch curvature features, PCA features, shape context features, average geodesic distance features, shape diameter features.
And 2, dividing Airplane, ant, chair, octopus, teddy, fish three-dimensional grid models in the Prlington model data set into a training data set and a test data set according to a proportion, wherein the number of each type of grid model is 20, namely, 19 models are randomly selected as the training data set to train the segmentation classifier, and the remaining 5 percent, namely, 1 model is used as the test data set to test the trained segmentation classifier because of Airplane, ant, chair, octopus, teddy, fish three-dimensional grid models in the Prlington model data set.
Step 3, an input feature vector P based on each type of three-dimensional grid model training data set obtained by dividing in step 2 is established, and the method specifically comprises the following steps: combining the surface patch features normalized in step 1 of models contained in each three-dimensional grid model training dataset to form an input feature vector P, wherein the input feature vector P with the number of samples being M is expressed as p= { (x) j ,t j ) J=1, 2, …, M }, where x j ＝(x j1 ,x j2 ,…,x jn ) T ∈R n Representing the extracted patch features for the three-dimensional mesh model, n is the attribute size of the input feature vector P, t j ＝(t j1 ,t j2 ,…,t jm ) T ∈R m Or as a matrixRepresenting the labeling of the three-dimensional mesh model.
Step 4, establishing a test feature vector P 'of each type of three-dimensional grid model test data set obtained based on the division in the step 2, wherein the test feature vector P' is specifically as follows: and (3) forming a test feature vector P' by the surface patch features of the three-dimensional grid model in the test data set after normalization processing in the step (1).
Step 5, setting the L parameter of the hidden layer neuron of the extreme learning machine model, which specifically comprises the following steps: setting the number L of hidden neurons to 300, and realizing a basic extreme learning algorithm by the steps 5.1 to 5.3:
step 5.1, randomly generating an input weight w i Offset b from hidden layer i ；
Step 5.2, according to the formulaCalculating a hidden layer output matrix H, wherein g (·) represents an activation function;
step 5.3, according to the formulaCalculating an output weight matrix, wherein H - Is an H generalized inverse matrix, matrixRepresenting the labeling of the three-dimensional mesh model.
Step 6, optimizing an extreme learning machine by adopting an ant lion optimization algorithm, and setting the maximum iteration number I in the ant lion optimization algorithm max 25 and population scale N is 20, the input feature vector P obtained based on the step 3 is used as input data, the input data is input into an ant lion optimized extreme learning machine model for training, an optimal segmentation classifier is obtained through training, and specific execution operation of segmentation test on the test feature vector P' obtained based on the step 4 is realized by the steps 6.1 to 6.6:
step 6.1, randomly initializing ant and ant lion populations according to a formulaRandomly initializing ant and ant lion populations to be NxLx (n+1), wherein the positions of each ant and ant lion individual in the populations represent parameter combinations f (w) to be optimized by the extreme learning machine i ,b i )；
Step 6.2, calculating the fitness value of each ant lion, and regarding the optimal as elite ant lion, wherein the steps from step 6.2.1 to step 6.2.7 are implemented:
step 6.2.1, dividing each ant-lion individual in the ant-lion population into an input weight w with the size of L multiplied by n i Implicit layer bias b of size L x 1 i ；
Step 6.2.2, according to the formulaCalculating a hidden layer output matrix H, wherein g (·) represents an activation function, and the activation function used in the method is an S-type function;
step 6.2.3, according to the formulaCalculating an output weight matrix, wherein H - Is an H generalized inverse matrix, matrixRepresenting the label marked by the three-dimensional grid model;
step 6.2.4, training the obtained segmentation classifier by the input feature vector P, wherein the segmentation classifier comprises an input weight, an implicit layer bias and an output weight, and predicting the label of each patch of the three-dimensional grid model of the test feature vector P';
step 6.2.5, comparing the predicted label with standard label, and counting the number of correct labels;
step 6.2.6, calculating the fitness value of the ant lion, namely testing the segmentation accuracy of the three-dimensional grid model of the feature vector P', wherein the segmentation accuracy is the ratio of the number of correctly marked grid surfaces to the total number of grid surfaces in the grid model;
step 6.2.7, comparing the fitness value of each individual in the ant lion population, sorting from big to small according to the fitness value, and assigning the optimal value to elite ant lion;
step 6.3, let t=1, enter the iterative loop, update the ant's position f (w i ,b i ) The method is realized by steps 6.3.1 to 6.3.3:
step 6.3.1, calculating the influence of the roulette strategy on the ant random walk to obtain the random walk selected by the roulette strategy in the t-th iteration
Step 6.3.2, calculating the influence of the elite ant lion on the random walk of the ants to obtain the random walk of the ants around the elite ant lion in the t-th iteration
Step 6.3.3, according to the formulaCalculating walking of each ant under the combined action of roulette wheel strategy and elite ant lion, and updating the position f (w i ,b i )；
Step 6.4, calculating the fitness value of each ant, which is realized by steps 6.4.1 to 6.4.6:
step 6.4.1, dividing each ant in the ant population into an input weight w with the size of L multiplied by n i Implicit layer bias b of size L x 1 i ；
Step 6.4.2, according to the formulaCalculating a hidden layer output matrix H, wherein g (·) represents an activation function, and the activation function used in the method is an S-type function;
step 6.4.3, according to the formulaCalculating the output weightValue matrix, wherein H - Is the H generalized inverse matrix;
step 6.4.4, a segmentation classifier is trained by the input feature vector P, the segmentation classifier comprises an input weight, an implicit layer bias and an output weight, and the label of each patch of the three-dimensional grid model of the test feature vector P' is predicted;
step 6.4.5, comparing the predicted label with the standard label, and counting the number of correct labels;
step 6.4.6, calculating the fitness value of the ant, namely testing the segmentation accuracy of the feature vector P' three-dimensional grid model, wherein the segmentation accuracy is the ratio of the number of correctly marked grid surfaces to the total number of grid surfaces in the grid model;
step 6.5, update elite lion, realized by step 6.5.1 to step 6.5.2:
step 6.5.1, the ant has better fitness value than ant lion, according to the formulaUpdate ant lion to captured ant position f (w i ,b i ) Otherwise, the position of the ant lion is kept unchanged;
step 6.5.2, the adaptation value of the ant lion is better than that of the elite lion, the position of the elite lion is updated, otherwise, the position of the elite lion is kept unchanged;
step 6.6, judging whether the ant lion optimization algorithm reaches the maximum iteration number I max If so, outputting an optimal fitness value corresponding to the elite ant lion, namely the segmentation accuracy and the position of the three-dimensional grid model of the test feature vector P', namely the parameter combination f (w i ,b i ) Otherwise, the iteration times t is added with 1, and the step 6.3 is skipped to continue execution until the maximum iteration times I of the algorithm are reached max ；
And 7, based on the optimal segmentation classifier obtained by training in the step 6, carrying out segmentation test on the test feature vector P 'obtained by training in the step 4 to obtain optimal segmentation accuracy of the three-dimensional grid model of the test feature vector P', generating a three-dimensional grid model segmentation result corresponding to the segmentation accuracy, and obtaining a visual result.
The beneficial effects of the invention are as follows:
compared with the prior art, the method integrates the steps 1 to 7, and has the remarkable advantages that: by utilizing the dual influence of elite and ant lion and roulette strategies in an ant lion optimization algorithm, iterative updating is carried out, and an optimal solution is assigned to the elite and ant lion, so that the input weight and hidden layer bias matrix randomly generated by an optimization extreme learning machine are improved, and a high-precision segmentation classifier is trained to test a Prins ston model. Compared with the work of the former, on Airplane, ant, chair, octopus, teddy and Fish6 model data sets, the model algorithm with the training of the number of patches between 20k and 30k takes about 1000 seconds, and the segmentation accuracy is higher than that of an unoptimized segmentation method, and the highest segmentation accuracy reaches 99.49%.
Drawings
FIG. 1 is an overall flow chart of the segmentation method of the present invention;
FIG. 2 is a basic extreme learning machine algorithm schematic;
FIG. 3 is a chart of convergence of different cluster sizes for an Airplane aircraft model;
FIG. 4 is a graph of convergence of Octopus models for different species sizes;
FIG. 5 is a graph of the results of the Airplane aircraft model segmentation;
FIG. 6 is a graph of the segmentation results of the Ant model;
FIG. 7 is a graph of the results of the Chair model segmentation;
FIG. 8 is a graph of the results of segmentation of an Octopus model of Octopus;
FIG. 9 is a graph of segmentation results of a Teddy bear model;
FIG. 10 is a graph of the results of the Fish model segmentation.
Detailed description of the preferred embodiments
The invention is described in further detail below with reference to the drawings and examples.
As shown in fig. 1, a flowchart of the three-dimensional mesh model segmentation method of the ant lion-based optimized extreme learning machine of the present invention can be summarized as steps 1 to 7:
step 1, for Airplane, ant, chair, octopus, teddy, fish three-dimensional grid models in a Prlington model dataset, extracting surface patch characteristics of each three-dimensional grid model, carrying out normalization processing, normalizing to be between [ -1,1] and eliminating dimension influence, wherein the surface patch characteristics comprise: patch curvature features, PCA features, shape context features, average geodesic distance features, shape diameter features.
And 2, dividing Airplane, ant, chair, octopus, teddy, fish three-dimensional grid models in the Prlington model data set into a training data set and a test data set according to a proportion, wherein the number of each type of grid model is 20, namely, 19 models are randomly selected as the training data set to train the segmentation classifier, and the remaining 5 percent, namely, 1 model is used as the test data set to test the trained segmentation classifier because of Airplane, ant, chair, octopus, teddy, fish three-dimensional grid models in the Prlington model data set.
Step 3, an input feature vector P based on each type of three-dimensional grid model training data set obtained by dividing in step 2 is established, and the method specifically comprises the following steps: combining the surface patch features normalized in step 1 of models contained in each three-dimensional grid model training dataset to form an input feature vector P, wherein the input feature vector P with the number of samples being M is expressed as p= { (x) j ,t j ) J=1, 2, …, M }, where x j ＝(x j1 ,x j2 ,…,x jn ) T ∈R n Representing the extracted patch features for the three-dimensional mesh model, n is the attribute size of the input feature vector P, t j ＝(t j1 ,t j2 ,…,t jm ) T ∈R m Or as a matrixRepresenting the labeling of the three-dimensional mesh model.
Step 4, establishing a test feature vector P 'of each type of three-dimensional grid model test data set obtained based on the division in the step 2, wherein the test feature vector P' is specifically as follows: and (3) forming a test feature vector P' by the surface patch features of the three-dimensional grid model in the test data set after normalization processing in the step (1).
Step 5, setting the L parameter of the hidden layer neuron of the extreme learning machine model, which specifically comprises the following steps: setting the number L of hidden layer neurons to 300, and realizing the basic extreme learning machine algorithm principle as shown in fig. 2 by steps 5.1 to 5.3:
step 5.1, randomly generating an input weight w i Offset b from hidden layer i ；
Step 5.2, according to the formulaCalculating a hidden layer output matrix H, wherein g (·) represents an activation function;
step 5.3, according to the formulaCalculating an output weight matrix, wherein H-is H generalized inverse matrix, and the matrixRepresenting the labeling of the three-dimensional mesh model.
Step 6, adopting an ant lion optimization algorithm to optimize an extreme learning machine, firstly selecting Airplane, octopus types of models in a Prins data set to perform test experiments, setting the population size N to be 10, 20 and 30 respectively, and counting the convergence of different species group scales of 50 iterations of the 2 models respectively as shown in fig. 3-4, wherein the running time of the segmentation method is doubled and increased along with the increase of the population scale, and simultaneously considering the running time and the segmentation accuracy, when the population scale N is set to be 20, the running time and the segmentation accuracy are well balanced, and when the iteration reaches about 25 th, the segmentation accuracy is basically stable, so that the maximum iteration number I in the ant lion optimization algorithm is set max 25 and population size N of 20, inputting the input feature vector P obtained based on the step 3 as input data into ant lion optimized extreme learning machine (ALO-ELM) model for training to obtain optimal segmentation classifier, and performing the step-based segmentation4, performing specific execution operation of the segmentation test on the obtained test feature vector P', wherein the specific execution operation is realized by steps 6.1 to 6.6:
step 6.1, randomly initializing ant and ant lion populations according to a formulaRandomly initializing ant and ant lion populations to be NxLx (n+1), wherein the positions of each ant and ant lion individual in the populations represent parameter combinations f (w) to be optimized by the extreme learning machine i ,b i )；
Step 6.2, calculating the fitness value of each ant lion, and regarding the optimal as elite ant lion, wherein the steps from step 6.2.1 to step 6.2.7 are implemented:
step 6.2.1, dividing each ant-lion individual in the ant-lion population into an input weight w with the size of L multiplied by n i Implicit layer bias b of size L x 1 i ；
Step 6.2.2, according to the formulaCalculating a hidden layer output matrix H, wherein g (·) represents an activation function, and the activation function used in the method is an S-type function;
step 6.2.3, according to the formulaCalculating an output weight matrix, wherein H - Is an H generalized inverse matrix, matrixRepresenting the label marked by the three-dimensional grid model;
step 6.2.4, training the obtained segmentation classifier by the input feature vector P, wherein the segmentation classifier comprises an input weight, an implicit layer bias and an output weight, and predicting the label of each patch of the three-dimensional grid model of the test feature vector P';
step 6.2.5, comparing the predicted label with standard label, and counting the number of correct labels;
step 6.2.6, calculating the fitness value of the ant lion, namely testing the segmentation accuracy of the three-dimensional grid model of the feature vector P', wherein the segmentation accuracy is the ratio of the number of correctly marked grid surfaces to the total number of grid surfaces in the grid model;
step 6.2.7, comparing the fitness value of each individual in the ant lion population, sorting from big to small according to the fitness value, and assigning the optimal value to elite ant lion;
step 6.3, let t=1, enter the iterative loop, update the ant's position f (w i ,b i ) The method is realized by steps 6.3.1 to 6.3.3:
step 6.3.1, calculating the influence of the roulette strategy on the ant random walk to obtain the random walk selected by the roulette strategy in the t-th iteration
Step 6.3.2, calculating the influence of the elite ant lion on the random walk of the ants to obtain the random walk of the ants around the elite ant lion in the t-th iteration
Step 6.3.3, according to the formulaCalculating walking of each ant under the combined action of roulette wheel strategy and elite ant lion, and updating the position f (w i ,b i )；
Step 6.4, calculating the fitness value of each ant, which is realized by steps 6.4.1 to 6.4.6:
step 6.4.1, dividing each ant in the ant population into an input weight w with the size of L multiplied by n i Implicit layer bias b of size L x 1 i ；
Step 6.4.2, according to the formulaCalculating hidden layer output matrix H, wherein g (·) represents the activation function, the presentThe activation function used in the method is an S-type function;
step 6.4.3, according to the formulaCalculating an output weight matrix, wherein H - Is the H generalized inverse matrix;
step 6.4.4, a segmentation classifier is trained by the input feature vector P, the segmentation classifier comprises an input weight, an implicit layer bias and an output weight, and the label of each patch of the three-dimensional grid model of the test feature vector P' is predicted;
step 6.4.5, comparing the predicted label with the standard label, and counting the number of correct labels;
step 6.4.6, calculating the fitness value of the ant, namely testing the segmentation accuracy of the feature vector P' three-dimensional grid model, wherein the segmentation accuracy is the ratio of the number of correctly marked grid surfaces to the total number of grid surfaces in the grid model;
step 6.5, update elite lion, realized by step 6.5.1 to step 6.5.2:
step 6.5.1, the ant has better fitness value than ant lion, according to the formulaUpdate ant lion to captured ant position f (w i ,b i ) Otherwise, the position of the ant lion is kept unchanged;
step 6.5.2, the adaptation value of the ant lion is better than that of the elite lion, the position of the elite lion is updated, otherwise, the position of the elite lion is kept unchanged;
step 6.6, judging whether the ant lion optimization algorithm reaches the maximum iteration number I max If so, outputting an optimal fitness value corresponding to the elite ant lion, namely the segmentation accuracy and the position of the three-dimensional grid model of the test feature vector P', namely the parameter combination f (w i ,b i ) Otherwise, the iteration times t is added with 1, and the step 6.3 is skipped to continue execution until the maximum iteration times I of the algorithm are reached max ；
And 7, based on the optimal segmentation classifier obtained by training in the step 6, performing segmentation test on the test feature vector P ' obtained by training in the step 4 to obtain optimal segmentation accuracy of the three-dimensional grid model of the test feature vector P ', generating a three-dimensional grid model segmentation result corresponding to the test feature vector P ', and obtaining a visual result, such as a Airplane, ant, chair, octopus, teddy, fish three-dimensional grid model segmentation result shown in fig. 5-10. The results of the visualizations in the present invention were compared with Chen X, golovinskiy A, funkhouser T.A benchmark for 3D mesh segmentation[J ]. ACM Transactions on Graphics,2009,28 (3): 1. The same procedure is used for visualizing the results in this article.
In order to show the advantages of the segmentation method, the experimental segmentation result of the invention is compared with a three-dimensional model segmentation method CGF [2014] based on basic extreme learning, a grid model segmentation method ATG [2015] based on a deep convolution network and a three-dimensional model segmentation method CEA [2017] based on rapid learning.
Table 1 comparison of the segmentation accuracy of the segmentation method of the present invention with other supervised based segmentation methods
As can be seen from Table 1, the segmentation method of the invention has greatly improved segmentation accuracy compared with CGF 2014 and also has improved segmentation accuracy compared with CEA 2017. In addition, although the segmentation method of the present invention has a segmentation result lower than ATG [2015], it takes about 1000 seconds to train the number of patches of 20k-30k in terms of time performance. The same number of patches shows an advantage over the method in ATG [2015] by taking about 4 hours. In general, the segmentation method of the invention applies the intelligent optimization idea ant lion optimization algorithm to the segmentation research of the grid model, adopts the basic model of the extreme learning machine, effectively avoids the problem of overlong training time of the deep neural network, and maintains higher segmentation accuracy than the former work, and the algorithm is also effective and feasible.
The above embodiments are not limited to the technical solution of the embodiments, and the embodiments may be combined with each other to form a new embodiment. The above embodiments are only for illustrating the technical solution of the present invention and are not limited thereto, and any modification or substitution without departing from the spirit and scope of the present invention should be covered in the scope of the technical solution of the present invention.
Claims (9)
1. A three-dimensional grid model segmentation method of an extreme learning machine based on ant lion optimization is characterized by comprising the following steps:
step 1, extracting surface patch characteristics of each three-dimensional grid model data aiming at Airplane, ant, chair, octopus, teddy, fish three-dimensional grid model data in a Prlington model data set, and carrying out normalization processing, wherein the surface patch characteristics comprise: the surface patch curvature characteristic, PCA characteristic, shape context characteristic, average geodesic distance characteristic and shape diameter characteristic are normalized to be between [ -1,1] and the dimension influence is eliminated;
step 2, dividing Airplane, ant, chair, octopus, teddy, fish three-dimensional grid model data in the Prins ston model data set into a training data set and a test data set according to proportion;
step 3, establishing an input feature vector P of each type of three-dimensional grid model training data set obtained based on the division in the step 2;
step 4, establishing a test feature vector P' of each type of three-dimensional grid model test data set obtained based on the division in the step 2;
step 5, setting the L parameter of the hidden layer neuron of the extreme learning machine model;
step 6, optimizing an extreme learning machine by adopting an ant lion optimization algorithm, and setting the maximum iteration number I in the ant lion optimization algorithm max And the size of the population scale N, taking the input feature vector P obtained based on the step 3 as input data, inputting the input feature vector P into an ant lion optimized extreme learning machine model for training to obtain an optimal segmentation classifier, and carrying out segmentation test on the test feature vector P' obtained based on the step 4;
and 7, based on the optimal segmentation classifier obtained by training in the step 6, carrying out segmentation test on the test feature vector P 'obtained by training in the step 4 to obtain optimal segmentation accuracy of the three-dimensional grid model of the test feature vector P', generating a three-dimensional grid model segmentation result corresponding to the segmentation accuracy, and obtaining a visual result.
2. The three-dimensional grid model segmentation method of the ant lion optimization-based extreme learning machine according to claim 1, wherein the three-dimensional grid model segmentation method is characterized in that: in the step 2, airplane, ant, chair, octopus, teddy, fish three-dimensional grid model data in the prinston model data set are respectively divided into a training data set and a test data set according to a proportion, specifically: the Airplane, ant, chair, octopus, teddy, fish three-dimensional grid model data in the Prins ston data set comprises 20 grid models, wherein 95% of the grid models in each type are randomly selected, namely 19 model data are used as training data sets for training the segmentation classifier, and the remaining 5% of the model data, namely 1 model data, are used as test data sets for checking the trained segmentation classifier.
3. The three-dimensional grid model segmentation method of the ant lion optimization-based extreme learning machine according to claim 1, wherein the three-dimensional grid model segmentation method is characterized in that: in the step 3, an input feature vector P of each type of three-dimensional grid model training data set obtained based on the division in the step 2 is established, specifically: combining the model data contained in the training data set of each type of three-dimensional grid model, normalizing the surface patch features in the step 1 to form an input feature vector P, wherein the input feature vector P with the number of samples M is expressed as P = { (x) j ,t j ) J=1, 2, …, M }, j representing the sample number, where x j ＝(x j1 ,x j2 ,…,x jn ) T ∈R n Xj represents the patch feature extracted by the jth three-dimensional grid model, xjn represents the nth patch of the jth sample, n is the patch size of the input feature vector P, t j ＝(t j1 ,t j2 ,…,t jm ) T ∈R m ，t jm Label data of the mth patch representing the jth sample, jm representing the jth sampleThe mth data of j samples, or as a matrixRepresenting the label marked by the three-dimensional grid model, T M T The vector representation of the tag data expressed as the mth sample, mxm represents the tag data matrix size constituted by the M tag data expressed as the M samples.
4. The three-dimensional grid model segmentation method of the ant lion optimization-based extreme learning machine according to claim 1, wherein the three-dimensional grid model segmentation method is characterized in that: in the step 4, a test feature vector P' of each type of three-dimensional grid model test data set obtained based on the division in the step 2 is established, specifically: and (3) forming a test feature vector P' by the surface patch features of the three-dimensional grid model in the test data set after normalization processing in the step (1).
5. The three-dimensional grid model segmentation method of the ant lion optimization-based extreme learning machine according to claim 1, wherein the three-dimensional grid model segmentation method is characterized in that: in the step 5, the magnitude of the hidden layer neuron L parameter of the extreme learning machine model is set, specifically: setting the number L of hidden layer neurons to 300; the basic algorithm of the extreme learning machine is realized by steps 5.1 to 5.3:
step 5.1, randomly generating an input weight w i Offset b from hidden layer i ；
Step 5.2, according to the formulaCalculating a hidden layer output matrix H, wherein g (·) represents an activation function; g (w) L ·x M +b L ) Is sample X M Is the output of the L hidden layer node, M x L represents that there are M samples and L hidden layers;
step 5.3, according to the formulaCalculating an output weight matrix, wherein H - Is the H generalized inverse matrix of the matrix,matrix->Representing the label marked by the three-dimensional grid model, T M T Vector representation of tag data represented as the mth sample, M x M represents the tag data matrix size of M tag data of the M samples, +.>The minimum output weight is calculated by the least square method.
6. The three-dimensional grid model segmentation method based on ant lion optimization extreme learning machine according to claim 5, wherein the three-dimensional grid model segmentation method is characterized in that: in the step 6, an ant lion optimization algorithm is adopted to optimize an extreme learning machine, and the maximum iteration number I in the ant lion optimization algorithm is set max 25 and population scale N is 20, the input feature vector P obtained based on the step 3 is used as input data, the input data is input into an ant lion optimized extreme learning machine model for training, an optimal segmentation classifier is obtained through training, and specific execution operation of segmentation test on the test feature vector P' obtained based on the step 4 is realized by the steps 6.1 to 6.6:
step 6.1, randomly initializing ant and ant lion populations according to a formulaRandomly initializing ant and ant lion populations to be NxLx (n+1), wherein the positions of each ant and ant lion individual in the populations represent parameter combinations f (w) to be optimized by the extreme learning machine i ,b i )，f(w i ,b i ) Representing weight input parameters and hidden layer bias parameters to be optimized;
step 6.2, calculating the fitness value of each ant lion, and regarding the optimal value as elite ant lion;
step 6.3, the iteration times t make t=1, enter the iteration loop, update the position f (w) i ,b i )；
Step 6.4, calculating the fitness value of each ant;
step 6.5, update elite lion, realized by step 6.5.1 to step 6.5.2:
step 6.5.1, the ant has better fitness value than ant lion, according to the formulaUpdate ant lion to captured ant position f (w i ,b i ) Otherwise, the position of the ant lion remains unchanged,/->Represents the position of the jth ant lion at the t iteration, < >>Representing the position of the ith ant at the t iteration;
step 6.5.2, the adaptation value of the ant lion is better than that of the elite lion, the position of the elite lion is updated, otherwise, the position of the elite lion is kept unchanged;
step 6.6, judging whether the ant lion optimization algorithm reaches the maximum iteration number I max If so, outputting an optimal fitness value corresponding to the elite ant lion, namely the segmentation accuracy and the position of the three-dimensional grid model of the test feature vector P', namely the parameter combination f (w i ,b i ) Otherwise, the iteration times t is added with 1, and the step 6.3 is skipped to continue execution until the maximum iteration times I of the algorithm are reached max 。
7. The three-dimensional grid model segmentation method based on ant lion optimization extreme learning machine according to claim 6, wherein the three-dimensional grid model segmentation method is characterized in that: in the step 6.2, the fitness value of each ant lion is calculated, and the optimal quality is regarded as the elite ant lion, and the steps from the step 6.2.1 to the step 6.2.7 are specifically realized:
step 6.2.1, dividing each ant-lion individual in the ant-lion population into an input weight w with the size of L multiplied by n i Implicit layer bias b of size L x 1 i ；
Step 6.2.2, according to the formulaCalculating a hidden layer output matrix H, wherein g (·) represents an activation function, and the activation function used in the method is an S-type function;
step 6.2.3, according to the formulaCalculating an output weight matrix, wherein H - Is an H generalized inverse matrix, matrixRepresenting the label marked by the three-dimensional grid model;
step 6.2.4, training the obtained segmentation classifier by the input feature vector P, wherein the segmentation classifier comprises an input weight, an implicit layer bias and an output weight, and predicting the label of each patch of the three-dimensional grid model of the test feature vector P';
step 6.2.5, comparing the predicted label with standard label, and counting the number of correct labels;
step 6.2.6, calculating the fitness value of the ant lion, namely testing the segmentation accuracy of the three-dimensional grid model of the feature vector P', wherein the segmentation accuracy is the ratio of the number of correctly marked grid surfaces to the total number of grid surfaces in the grid model;
and 6.2.7, comparing the fitness value of each individual in the ant lion population, sorting from large to small according to the fitness value, and assigning the optimal value to the elite ant lion.
8. The three-dimensional grid model segmentation method based on ant lion optimization extreme learning machine according to claim 6, wherein the three-dimensional grid model segmentation method is characterized in that: in the step 6.3, let t=1, enter an iterative loop, update the position f (w i ,b i ) Specifically, the method is realized by steps 6.3.1 to 6.3.3:
step 6.3.1, calculating roulette strategy pair antInfluence of ant random walk to obtain random walk selected by roulette strategy at t-th iteration Is randomly walked selected by the roulette strategy at the t-th iteration;
step 6.3.2, calculating the influence of the elite ant lion on the random walk of the ants to obtain the random walk of the ants around the elite ant lion in the t-th iteration Is a random walk of ants around elite ant lions at the t-th iteration;
step 6.3.3, according to the formulaRepresenting the positions of ants in the ant lion optimization algorithm, calculating the walking of each ant under the combined action of the roulette wheel strategy and elite ant lion, and updating the positions f (w i ,b i )。
9. The three-dimensional grid model segmentation method based on ant lion optimization extreme learning machine according to claim 6, wherein the three-dimensional grid model segmentation method is characterized in that: in the step 6.4, the fitness value of each ant is calculated, and the method is specifically realized by the steps 6.4.1 to 6.4.6:
step 6.4.1, dividing each ant individual in the ant population into an input weight wi with the size of L multiplied by n and an hidden layer bias bi with the size of L multiplied by 1;
step 6.4.2, according to the formulaCalculating hidden layer output matrix H, whichThe g (-) in the method represents an activation function, and the activation function used by the method is an S-type function;
step 6.4.3, according to the formulaCalculating an output weight matrix, wherein H-is an H generalized inverse matrix;
step 6.4.4, a segmentation classifier is trained by the input feature vector P, the segmentation classifier comprises an input weight, an implicit layer bias and an output weight, and the label of each patch of the three-dimensional grid model of the test feature vector P' is predicted;
step 6.4.5, comparing the predicted label with the standard label, and counting the number of correct labels;
in step 6.4.6, the fitness value of the ant, that is, the segmentation accuracy of the three-dimensional mesh model of the test feature vector P', is calculated, where the segmentation accuracy is the ratio of the number of correctly marked mesh surfaces to the total number of mesh surfaces in the mesh model.
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 CN110288606A (en) | 2019-09-27 |
| CN110288606B true 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) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
Citations (7)
| 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 |
| 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 |
| 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 |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105550744A (en) * | 2015-12-06 | 2016-05-04 | 北京工业大学 | Nerve network clustering method based on iteration |
| US20180240018A1 (en) * | 2016-05-19 | 2018-08-23 | Jiangnan University | Improved extreme learning machine method based on artificial bee colony optimization |
-
2019
- 2019-06-28 CN CN201910576352.4A patent/CN110288606B/en active Active
Patent Citations (7)
| 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 |
| 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 |
|---|
| An Ant-Lion Optimizer-Trained Artificial Neural Network System for Chaotic Electroencephalogram (EEG) Prediction;Utku Kose;《Applied Sciences》;第1-32页 * |
| The Ant Lion Optimizer;Mirjalili S;《Advances in Engineering Software》;第80-98页 * |
| 一种基于蚁群优化的网格分割方法;张耀楠 等;《计算机工程》;第44卷(第2期);第277-281页 * |
| 三维形状分割和标注的快速学习方法;李红岩;《计算机工程与应用》;第53卷(第11期);第211-216页 * |
| 改进人工蜂群优化的极限学习机;毛羽 等;《传感器与微系统》;第37卷(第4期);第116-120页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110288606A (en) | 2019-09-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110689086B (en) | Semi-supervised high-resolution remote sensing image scene classification method based on generating countermeasure network | |
| WO2022178977A1 (en) | Unsupervised data dimensionality reduction method based on adaptive nearest neighbor graph embedding | |
| Yang et al. | Deep learning for the classification of lung nodules | |
| CN105809672B (en) | A kind of image multiple target collaboration dividing method constrained based on super-pixel and structuring | |
| CN113408605B (en) | Hyperspectral image semi-supervised classification method based on small sample learning | |
| CN103593855B (en) | The image partition method of cluster is estimated based on particle group optimizing and space length | |
| CN111914728B (en) | Hyperspectral remote sensing image semi-supervised classification method and device and storage medium | |
| CN107203785A (en) | Multipath Gaussian kernel Fuzzy c-Means Clustering Algorithm | |
| CN109902736A (en) | A kind of Lung neoplasm image classification method indicated based on autocoder construction feature | |
| CN112396587B (en) | Method for detecting congestion degree in bus compartment based on collaborative training and density map | |
| CN112418330A (en) | Improved SSD (solid State drive) -based high-precision detection method for small target object | |
| CN115661550B (en) | Graph data category unbalanced classification method and device based on generation of countermeasure network | |
| Islam et al. | InceptB: a CNN based classification approach for recognizing traditional bengali games | |
| CN113157957A (en) | Attribute graph document clustering method based on graph convolution neural network | |
| CN107153837A (en) | Depth combination K means and PSO clustering method | |
| CN111524140B (en) | Medical image semantic segmentation method based on CNN and random forest method | |
| CN108364073A (en) | A kind of Multi-label learning method | |
| CN114822823B (en) | Tumor fine classification system based on cloud computing and artificial intelligence fusion multi-dimensional medical data | |
| CN110288606B (en) | Three-dimensional grid model segmentation method of extreme learning machine based on ant lion optimization | |
| CN110175631A (en) | A kind of multiple view clustering method based on common Learning Subspaces structure and cluster oriental matrix | |
| CN109886303A (en) | A kind of TrAdaboost sample migration aviation image classification method based on particle group optimizing | |
| CN108921853A (en) | Image partition method based on super-pixel and clustering of immunity sparse spectrums | |
| CN115273645B (en) | Map making method for automatically clustering indoor surface elements | |
| CN106600100A (en) | Reason analysis method for danger source based on weighted multi-population particle swarm optimization (PSO) | |
| CN111353525A (en) | Modeling and missing value filling method for unbalanced incomplete data set |
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 |