CN109063271A - A kind of three-dimensional CAD model dividing method and device based on the learning machine that transfinites - Google Patents
A kind of three-dimensional CAD model dividing method and device based on the learning machine that transfinites Download PDFInfo
- Publication number
- CN109063271A CN109063271A CN201810758421.9A CN201810758421A CN109063271A CN 109063271 A CN109063271 A CN 109063271A CN 201810758421 A CN201810758421 A CN 201810758421A CN 109063271 A CN109063271 A CN 109063271A
- Authority
- CN
- China
- Prior art keywords
- attribute
- cad model
- dimensional cad
- learning machine
- adjacent label
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of three-dimensional CAD model dividing methods and device based on the learning machine that transfinites, and calculate feature corresponding to each face of three-dimensional CAD model and describe operator；Feature based on all faces describes operator, and trained and test is transfinited learning machine；Each face of three-dimensional CAD model is classified and marked using learning machine is transfinited；Based on classification results, the adjacent label figure of attribute of three-dimensional CAD model is constructed；The adjacent label figure of attribute is split；Optimization is merged to the adjacent label figure of attribute after segmentation using the maximum cohesion degree of the adjacent label figure segmentation of attribute as objective function, obtains multiple regional areas.The present invention classifies to the plane of three-dimensional CAD model, concave surface, convex surface using learning machine is transfinited, and three-dimensional CAD model is expressed with the adjacent label figure of attribute, then according to the adjacent label figure of the corresponding attribute of three-dimensional CAD model, is split and optimizes.
Description
Technical field
The present invention relates to three-dimensional CAD models to divide field, and in particular to a kind of three-dimensional CAD model based on the learning machine that transfinites
Dividing method and device.
Background technique
The segmentation of CAD model and mark are the basic projects that threedimensional model feature understands.Thus CAD model segmentation is to three
Key job before dimension module feature subsequent processing, traditional surface segmentation method is primarily directed to grid model, three-dimensional CAD mould
The expression of type is usually to use B-rep B reps.How CAD model is divided into the partial zones with certain engineering significance
Domain is integrated into urgent problem to be solved.
Conventional CAD model segmentation is by the way of artificial, and mode manually carries out the segmentation work effect of CAD model
Rate is low, low precision.As the fast pace in machine learning field develops, some new classification methods based on machine learning are suggested
Come.Currently, mainly having using machine learning to the method that CAD model is divided: supervised learning dividing method and unsupervised learning
Dividing method.Model based on supervised learning be segmented in Princeton University's model segmentation assessment collection it is best the result is that
94%, but the training speed of supervised learning dividing method is lower.In unsupervised learning dividing method, researchers are not using
With clustering method clustered, be then split and mark.In order to accelerate the calculating of cluster, such methods are usually first
Over-segmentation is carried out for threedimensional model, it, can greatly improvement method then in the extraction and further cluster for carrying out feature
Speed, but the effect of final segmentation result heavy dependence and over-segmentation.
In conclusion it is too long for the training time in the prior art, Generalization accuracy is low and is easily trapped into local minimum
The problem of, still lack effective solution scheme.
Summary of the invention
In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides a kind of three-dimensional CAD moulds based on the learning machine that transfinites
Type dividing method and device, using transfinite learning machine training neural network to the plane of three-dimensional CAD model, concave surface, convex surface into
Row classification, using the B-rep representation of three-dimensional CAD model as input source, by three-dimensional CAD model with the adjacent label figure of attribute into
Row expression is split and optimizes then according to the adjacent label figure of the corresponding attribute of three-dimensional CAD model.
The technical scheme adopted by the invention is that:
The first object of the present invention is to provide a kind of three-dimensional CAD model dividing method based on the learning machine classifier that transfinites,
Method includes the following steps:
It calculates feature corresponding to each face of three-dimensional CAD model and describes operator；
Feature based on all faces describes operator, and trained and test is transfinited learning machine；
Each face of three-dimensional CAD model is classified and marked using learning machine is transfinited；
Based on classification results, the adjacent label figure of attribute of three-dimensional CAD model is constructed；
The adjacent label figure of attribute is split；
Maximum cohesion degree using the adjacent label figure segmentation of attribute be objective function it is adjacent to the attribute after segmentation mark figure into
Row merges optimization, obtains multiple regional areas.
Further, it includes based on principal component point that feature corresponding to each face of the three-dimensional CAD model, which describes operator,
Feature, face curvature feature and the shape characteristics of diameters of analysis.
Further, the training and test method of the learning machine that transfinites are as follows:
Place is normalized in the feature based on principal component analysis, face curvature feature and shape characteristics of diameters in all faces
Reason, obtains a vector, as the input feature value for the learning machine classifier training that transfinites, is input in the learning machine that transfinites and carries out
Training；
The transfinite hidden layer node number of learning machine, training pattern number and neuron emergency function, removal of selection is transfinited
Original weight term in learning machine.
Further, described the step of each face of three-dimensional CAD model is classified and marked using the learning machine that transfinites
Include:
Several neuron probability values in each face of three-dimensional CAD model are calculated using the learning machine that transfinites；
Several neuron probability values in each face are normalized, the label probability value in the face is obtained；
Classified using the label probability value in each face to each face of three-dimensional CAD model, distinguishes three-dimensional CAD model
Plane, convex surface, concave surface.
Further, the construction method of the adjacent label figure of the attribute of the three-dimensional CAD model are as follows:
The data structure of the adjacent label figure of defined attribute, including adjacency and concavity and convexity；
It traverses each face of threedimensional model and extracts all properties in each face, the corresponding section of the adjacent label figure of creation attribute
Point；
Identify the syntople between each face of threedimensional model, the side of the adjacent label figure of creation attribute.
Further, the dividing method of the adjacent label figure of the attribute are as follows:
It is split according to the attribute of the adjacent label figure interior joint of attribute and line label figure G adjacent to attribute, if obtaining
Dry regional area subgraph, constitutes regional area collection S；
The line between each node and node in regional area collection S is deleted in the adjacent label figure G of dependence, is obtained
The adjacent label figure G ' of new attribute；
If new attribute linkage flag figure G ' is sky, show to have been completed in the adjacent label figure of attribute all nodes with
And the segmentation of line；
If in new attribute linkage flag figure G ' including the subgraph of Combination node, divide again according to first identification segmentation caviton figure
The principle for cutting convex portion figure divides the attribute linkage flag figure G ' comprising Combination node again, until obtaining new attribute
Adjacent label figure is empty.
It further, further include the cohesion degree and region couples of setting regions cohesion degree, the adjacent label figure segmentation of attribute
The step of spending, the region cohesion degree are the average degree of each node in the adjacent label figure of attribute；The attribute linkage flag figure
The cohesion degree of segmentation is the mean value of the corresponding each regional area cohesion degree of three-dimensional CAD model；The region couples degree is partial zones
Company of the node in any two regional area subgraph in the corresponding attribute linkage flag figure of three-dimensional CAD model in the collection S of domain
Line.
Further, the adjacent label figure of attribute after described pair of segmentation merges optimization method are as follows:
According to the cohesion degree expression formula of the adjacent label figure segmentation of attribute, the interior of the regional area collection S obtained after segmentation is calculated
Poly- degree, obtains new regional area collection S ' after update；
Each regional area subgraph G in localized region collection S 'i' analyzed, several alternative merging subgraphs are obtained,
Constitute alternative set A；
It is selected from alternative set A and GiThe maximum subgraph of ' degree of coupling merges processing；
According to the cohesion degree of the adjacent label figure segmentation of attribute, the cohesion degree of regional area collection S ' after segmentation is calculated, will be divided
The cohesion degree of preceding regional area collection S is compared with the cohesion degree of the regional area collection S ' after segmentation, if regional area collection S '
Cohesion degree be less than regional area collection S cohesion degree, then export regional area collection S '.
Further, each regional area subgraph G in the localized region collection S 'i' the method analyzed are as follows:
By regional area subgraph G each in regional area collection S 'i' attribute analyzed；
Judge whether there is the node of subgraph and the concavity of line and regional area subgraph Gi' consistent；
The node of subgraph and the concavity of connection and regional area subgraph G if it existsi' identical, then using the subgraph as standby
The merging subgraph of choosing.
The second object of the present invention is to provide a kind of three-dimensional CAD model segmenting device based on the learning machine classifier that transfinites,
The device include memory, processor and storage on a memory and the computer program that can run on a processor, the place
Reason device realizes following steps when executing described program, comprising:
It calculates feature corresponding to each face of three-dimensional CAD model and describes operator；
Feature based on all faces describes operator, and trained and test is transfinited learning machine；
Each face of three-dimensional CAD model is classified and marked using learning machine is transfinited；
Based on classification results, the adjacent label figure of attribute of three-dimensional CAD model is constructed；
The adjacent label figure of attribute is split；
Maximum cohesion degree using the adjacent label figure segmentation of attribute be objective function it is adjacent to the attribute after segmentation mark figure into
Row merges optimization, obtains multiple regional areas.
Compared with prior art, the beneficial effects of the present invention are:
(1) present invention extracts feature for each face of three-dimensional CAD model, trains the learning machine that transfinites later, and passing through should
The segmentation of the three-dimensional CAD model of the learning machine that transfinites prediction input and mark can be obtained using the learning machine that transfinites as classifier
Quick training and test speed；
(2) compared with the BP neural network of traditional application error gradient decline learning strategy, what the present invention used transfinites
Quickly, Generalization accuracy is high, and will not fall into local minimum for learning machine pace of learning, can use a variety of excitation functions.
Detailed description of the invention
The accompanying drawings constituting a part of this application is used to provide further understanding of the present application, and the application's shows
Meaning property embodiment and its explanation are not constituted an undue limitation on the present application for explaining the application.
Fig. 1 is the three-dimensional CAD model dividing method flow chart one based on the learning machine that transfinites；
Fig. 2 is the three-dimensional CAD model dividing method flowchart 2 based on the learning machine that transfinites；
Fig. 3 is three-dimensional CAD model schematic diagram；
Fig. 4 is that the adjacent label diagram of attribute of three-dimensional CAD model is intended to；
Fig. 5 is the adjacent label figure segmentation exemplary diagram of attribute.
Specific embodiment
The invention will be further described with embodiment with reference to the accompanying drawing.
It is noted that following detailed description is all illustrative, it is intended to provide further instruction to the application.Unless another
It indicates, all technical and scientific terms used herein has usual with the application person of an ordinary skill in the technical field
The identical meanings of understanding.
It should be noted that term used herein above is merely to describe specific embodiment, and be not intended to restricted root
According to the illustrative embodiments of the application.As used herein, unless the context clearly indicates otherwise, otherwise singular
Also it is intended to include plural form, additionally, it should be understood that, when in the present specification using term "comprising" and/or " packet
Include " when, indicate existing characteristics, step, operation, device, component and/or their combination.
As background technique is introduced, the training time is too long in the prior art, Generalization accuracy is low and is easily trapped into office
The deficiency of portion's minimum value, in order to solve technical problem as above, present applicant proposes a kind of three-dimensional CADs based on the learning machine that transfinites
Model dividing method.
Embodiment 1
In a kind of typical embodiment of the application, as shown in Figure 1, providing a kind of three-dimensional based on the learning machine that transfinites
CAD model dividing method, method includes the following steps:
Step 101: calculating feature corresponding to each face of three-dimensional CAD model and describe operator；
Step 102: the feature based on all faces describes operator, and trained and test is transfinited learning machine；
Step 103: each face of three-dimensional CAD model being classified and marked using learning machine is transfinited；
Step 104: being based on classification results, construct the adjacent label figure of attribute of three-dimensional CAD model；
Step 105: the adjacent label figure of attribute is split；
Step 106: the cohesion degree and region couples degree of the adjacent label figure segmentation of definition region cohesion degree, attribute, to belong to
Property adjacent label figure segmentation maximum cohesion degree be that objective function merges optimization to the adjacent label figure of attribute after segmentation, obtain
To multiple regional areas.
The embodiment of the invention provides a kind of three-dimensional CAD model dividing methods based on the learning machine that transfinites, for three-dimensional CAD
Feature is extracted in each face of model, trains the learning machine that transfinites later, and pass through the three-dimensional CAD mould of learning machine prediction input of transfiniting
The segmentation of type and mark can obtain quick training and test speed using the learning machine that transfinites as classifier.
Embodiment 2
In order to make those skilled in the art be better understood by the present invention, a more detailed embodiment is set forth below,
As shown in Fig. 2, the embodiment of the invention provides a kind of three-dimensional CAD model dividing method based on the learning machine that transfinites, this method packet
Include following steps:
Step 201: the feature for calculating each three-dimensional CAD model describes operator.
Corresponding feature is calculated to each face of the three-dimensional CAD model of the selection and describes operator, these features are retouched
The attribute in each face can sufficiently be reflected by stating operator, and comprehensive these types feature, which can utilize, transfinites learning machine to three-dimensional CAD mould
The corresponding face of type distinguishes (plane, convex surface, concave surface).
The corresponding feature in each face describe operator include feature based on principal component analysis, face curvature feature (principal curvatures,
Minimum curvature, maximum curvature), shape characteristics of diameters.
The calculation method of the feature based on principal component analysis are as follows:
Fp=a1i*ZX1+a2i*ZX2+……+api*ZXp
Wherein, a1i、a2i、……、api(i=1 ..., m) be feature corresponding to the characteristic value of the covariance matrix Σ of X to
Amount, ZX1、ZX2、……、ZXpValue of the original variable Jing Guo standardization, m is principal component number, p be corresponding feature to
The number of amount；Often there are the dimension of index differences in calculating process, so the influence of dimension is first eliminated before the computation, and
Initial data is standardized.
A=(aij) p × m=(a1, a2... am)；
Rai=λiai,
A is corresponding feature vector after standardization；R is correlation matrix, λi、aiBe corresponding characteristic value and
Unit character vector, wherein λ1≥λ2≥…≥λp≥0。
The face curvature feature calculation method are as follows:
Curved surface available parameter equation indicates
P (u, v)=(x (u, v), y (u, v), z (u, v))
Wherein, x (u, v) is x in u, the component in the direction v；Y (u, v) is y in u, the component in the direction v；Z (u, v) is z in u, v
The component in direction；P (u, v) indicates the parametric equation of curved surface, the curved surface indicated with parametric equation.
It is then as follows about the first directional derivative on the direction u and v:
Pu(u, v)=(xu(u,v),yu(u,v),zu(u,v))
Pv(u, v)=(xv(u,v),yv(u,v),zv(u,v))
It is assumed that:
E=pu·pu
F=pv·pu=pu·pv
G=pv·pv
An available symmetric determinant:
PuWith PvA partial derivative combination can indicate are as follows:
ξpu+ηpv
The secondary product of combination are as follows:
(ξpu+ηpv)·(ξpu+ηpv)=E ξ2+2Fξη+Gη2
Wherein, ξ is expression puCoefficient；η is expression pvCoefficient.puWith pvA partial derivative combination can indicate are as follows:
ξpu+ηpv
The present invention, which describes operator by these features, can sufficiently reflect the attribute in each face, comprehensive these types feature energy
It is enough that (plane, convex surface, concave surface) is distinguished to the corresponding face of three-dimensional CAD model using the learning machine that transfinites.
Step 202: training and test the learning machine stage of transfiniting.
The training and test method of the learning machine that transfinites are as follows:
By the feature based on principal component analysis, face curvature feature, shape diameter in each face of selected three-dimensional CAD model
Feature is normalized, these operator tuples are become a vector, and the input as the learning machine classifier training that transfinites is special
Vector is levied, the learning machine that transfinites is input to and is trained；
The parameters such as hidden layer node number, training pattern number and neuron the emergency function to the learning machine that transfinites are selected
It selects.
The learning machine that transfinites can automatically learn the corresponding weight of neuron, therefore it can be to each face of input model
Attributive character is automatically selected.
The present invention can obtain quick training and test speed using learning machine is transfinited as classifier.
Step 203: using the neural network for learning machine training of transfiniting, model being carried out to each face of three-dimensional CAD model
Plane, concave surface, convex surface are classified and are marked.
The label probability value in each face of three-dimensional CAD model is calculated using the learning machine that transfinites, the label probability value utilized is to three
Classify in each face of Victoria C AD model.Its concrete methods of realizing are as follows:
The present invention regards threedimensional model to be split as a figure, is denoted as G={ V, E }, and the node V in figure corresponds to the mould
Each face of type, it is syntople in a model that the arc { u, v } in figure, which represents face u and v,.Remember the segmentation tag number n of the model.
Training and test process by ELM classifier, each available n neuron probability value of face u in model to be split.Root
According to the ELM theories of learning, face label luIt is to be obtained by the corresponding label of maximum value of this n neuron probability value.We are opposite u
N neuron probability value normalization, available its label probability value p (lu| u), thus to threedimensional model face preferably into
Row classification.
Using the label probability value in each face of three-dimensional CAD model by each face of the three-dimensional CAD model of selection with being divided
Class distinguishes CAD model plane, convex surface, concave surface.
The present invention is using learning machine method one neural network of training that transfinites come the plane, concave surface, convex to three-dimensional CAD model
Face is classified, and without manually classifying, is accelerated the efficiency and accuracy of the manually classification to the face CAD, is simplified
The step of CAD model is divided, the learning machine that transfinites are based on the single layer feedforward neural network promoted to realize function, because transfiniting
Learning machine only needs the weight for generating neuron at random and only least square is needed to carry out adjustment parameter, without the behaviour using iteration
Make, therefore the training speed for the learning machine that transfinites is very fast, and the effect tested is also fine.
Step 204: establishing the adjacent label figure of attribute of selected three-dimensional CAD model.
The construction method of the adjacent label figure of the attribute of the three-dimensional CAD model are as follows:
(1) data structure, including adjacency, concavity and convexity etc. of adjacent label (AALG) figure of defined attribute；
(2) it traverses each face of three-dimensional CAD model and extracts each attribute in each face, the adjacent label figure of creation attribute
Corresponding node；
(3) syntople between each face of three-dimensional CAD model, the side of the adjacent label figure of creation attribute are identified.
Three-dimensional CAD model is as shown in figure 3, the adjacent label of the attribute of three-dimensional CAD model is schemed as shown in figure 4, wherein V indicates section
The set of point, each node have unique face in model to be corresponding to it；E indicates the set of line, and every line has model
In unique side be corresponding to it；CV={ 0,1,2 } is the concavity label sets of node, indicates the concavity in corresponding face, wherein 0
Indicate convexity, 1 indicates concavity, and 2 indicate Combination；LE={+,-, 0 } is the label sets of line, indicates the concavity of corresponding sides,
Wherein+chimb is represented ,-concave edge is represented, 0 indicates trimming.
Connection side between face and face can be differentiated that be divided into chimb, concave edge is cut according to the outer angle between curved surface
Side.As soon as if face belongs to convex surface, either plane and its connection are when being not admitted to concave edge and undercut, this kind of node
As the convex node of figure, if a face belongs to concave curved surface either plane and its connection side and is not admitted to chimb and convex cuts
Side, this kind of node just become the concave section point of figure.If a node meets convex (recessed) curved surface and its company for the curved surface that it is indicated
There are recessed (convex) side or recessed (convex) trimming in edge fit.
The present invention establishes the adjacent label figure of attribute of three-dimensional CAD model, and the adjacent label figure of each attribute corresponds to one
The continuous regional area of concavity, divides threedimensional model using attribute adjacent map, is more intuitively partitioned into cohesion degree more
It gets well, more the regional area of engineering significance.
Step 204: three-dimensional CAD model is split.
First according to attributes such as concavities of adjacent label figure (AALG) interior joint of attribute and line to the adjacent label of attribute
Figure (AALG) is split, then respectively the cohesion degree of adjacent label figure (AALG) segmentation of definition region cohesion degree, attribute and
Region couples degree merges optimization as target using the maximum cohesion degree of adjacent label figure (AALG) segmentation of attribute, is partitioned into more
Satisfactory multiple regional areas.
Step 2041: the adjacent label figure of attribute is split.
The dividing method of the adjacent label figure of the attribute are as follows:
Label figure G adjacent to the attribute with determining concavity and convexity is split, and the result of segmentation is deposited to regional area
Collect S；Then each node in regional area collection S and the line between them are deleted in the adjacent label figure G of dependence, will be worth
It is assigned to the adjacent label figure G ' of new attribute；If the adjacent label figure G ' of this stylish attribute is sky, illustrate to have been completed attribute
The segmentation of all nodes and line in adjacent label figure G, otherwise the adjacent label figure G ' of new attribute includes Combination node.
Divide the principle of convex portion figure again to the subgraph G ' progress comprising Combination node according still further to first identification segmentation caviton figure
Further segmentation terminates until the adjacent label figure G of all attributes all has clear concavity and convexity.
Step 2042: the cohesion degree and region couples of adjacent label figure (AALG) segmentation of definition region cohesion degree, attribute
Degree.
The method of adjacent label figure (AALG) segmentation of attribute is also easy to produce a considerable amount of " acnodes " or the seldom son of number of nodes
Figure, causes cohesion between the internal node of region to be spent greatly, therefore model area segmentation effect is still undesirable.
Correlation degree in the adjacent label figure (AALG) of attribute corresponding to region between each node, it is related between node
Being combined line indicates relationship, and in this case, the average degree that region cohesion degree can use each node of subgraph goes to define.
The cohesion degree of adjacent label figure (AALG) segmentation of attribute: the corresponding each regional area cohesion degree of three-dimensional CAD model
Mean value.
Region couples degree: any two partial zones after attribute adjacent label figure (AALG) is divided in regional area collection S are set
Domain subgraph, line of the node in the adjacent label figure of the corresponding attribute of three-dimensional CAD model in two regional area subgraphs reflect
Degree of coupling between two subgraphs.
Step 2043: merging optimization.
Annexable condition is as follows: for each subgraph of three-dimensional CAD model, if after the completion of the merging of adjacent subgraph
It is still region subgraph, then the two subgraphs can merge, as shown in Figure 5.
AALG segmentation optimization merging method based on cohesion degree are as follows:
Input: the primary segmentation result S of AALG.
Output: the optimization amalgamation result S ' of segmentation.
(1) S ' ← S is calculated the cohesion degree of S, obtains S ' after update according to the cohesion degree expression formula that AALG is divided.
The cohesion degree expression formula of AALG segmentation are as follows:
In formula, Gi' induced subgraph for scheming S ' is marked for attribute adjoining；V is the node in corresponding subgraph；vjFor node j；|
Gi′ vj| it is subgraph Gi' number of nodes.
(2) to each subgraph G of S 'i', by subgraph G each in S 'i' analyzed, if there is subgraph, this subgraph
The concavity and G of node and linei' as, then the merging subgraph by this subgraph alternately, by all alternative sons
Set of graphs together, referred to as alternative set A；Then it selects from alternative set A and GiThe maximum subgraph of ' degree of coupling merges
Processing.
The evaluation of the cutting operation and region cohesion degree and the degree of coupling of AALG is completed to each subgraph of treated S '.
The definition of region couples degree: any two region G in AALG segmentation S ' is seti', Gj', the node in two subgraphs exists
Line in the adjacent label figure G of attribute reflects the degree of coupling degree between two subgraphs.
The expression formula of region couples degree: f (Gi′,Gj')=ink (Gi′,Gj′)
What Link () was indicated is the wiring quantity between two subgraph interior joints in G, more reflecting regional of session number
Degree of coupling is just higher between figure.
When carrying out subgraph merging, if there is multiple alternative merging subgraphs for meeting the convex-concave condition of continuity, then preferentially examine
Worry is merged with the maximum subgraph of region couples degree, because the degree of coupling is bigger between subgraph, illustrates that its interregional association is closer,
It is bigger that they belong to a possibility that the same area.
(3) according to the cohesion degree for defining AALG segmentation, the cohesion degree of S ' after segmentation is calculated, the cohesion degree of S and S ' is carried out
Compare, if the cohesion degree of S ' is greater than the cohesion degree of S, step (1) is gone to, otherwise, is exported as S ', end algorithm.
In order to further increase the effect of CAD model segmentation, three-dimensional CAD model segmentation problem is converted basis by the present invention
The node of figure and the even attribute on side extract B-rep form three-dimensional CAD model to improve the effect of three-dimensional CAD model segmentation
Attribute, be the adjacent label figure of attribute by model conversion；Then the concavity according to figure interior joint and line tentatively divides model
It is segmented into the continuous regional area of concavity, then is up to target to divide cohesion degree, combinable area is carried out based on region couples degree
The selection and optimization in domain merges；Using figure node and even the attribute on side, relations problems by the corresponding attribute of three-dimensional CAD model
The problem of adjacent label figure is divided into several subgraphs, this drawing of seeds are that concavity and convexity is continuous and mutually disjoint induced subgraph.
The embodiment of the present invention proposes a kind of three-dimensional CAD model dividing method based on the learning machine that transfinites, and calculates each three
The feature of Victoria C AD model describes operator, these operator tuples is become a vector, as the defeated of learning machine classifier training that transfinite
Enter feature vector, later, the learning machine that transfinites is trained, and input model is carried out using the trained learning machine classifier that transfinites
Classification and mark, are expressed as information input source with the B-rep of CAD model, first by the three-dimensional CAD model adjacent label figure of attribute
To indicate；Then according to the adjacent label figure of attribute, with region maximum cohesion degree of segmentation etc. for objective function, to CAD model into
Row segmentation collection optimization merges.
Above-mentioned, although the foregoing specific embodiments of the present invention is described with reference to the accompanying drawings, not protects model to the present invention
The limitation enclosed, those skilled in the art should understand that, based on the technical solutions of the present invention, those skilled in the art are not
Need to make the creative labor the various modifications or changes that can be made still within protection scope of the present invention.
Claims (10)
1. a kind of three-dimensional CAD model dividing method based on the learning machine classifier that transfinites, characterized in that the following steps are included:
It calculates feature corresponding to each face of three-dimensional CAD model and describes operator；
Feature based on all faces describes operator, and trained and test is transfinited learning machine；
Each face of three-dimensional CAD model is classified and marked using learning machine is transfinited；
Based on classification results, the adjacent label figure of attribute of three-dimensional CAD model is constructed；
The adjacent label figure of attribute is split；
The adjacent label figure of attribute after segmentation is closed using the maximum cohesion degree of the adjacent label figure segmentation of attribute as objective function
And optimize, obtain multiple regional areas.
2. the three-dimensional CAD model dividing method according to claim 1 based on the learning machine classifier that transfinites, characterized in that
It includes that feature, face curvature based on principal component analysis are special that feature corresponding to each face of the three-dimensional CAD model, which describes operator,
Shape of seeking peace characteristics of diameters.
3. the three-dimensional CAD model dividing method according to claim 1 based on the learning machine classifier that transfinites, characterized in that
The training and test method of the learning machine that transfinites are as follows:
The feature based on principal component analysis, face curvature feature and shape characteristics of diameters in all faces are normalized, obtained
It is input in the learning machine that transfinites and is trained as the input feature value for the learning machine classifier training that transfinites to a vector；
Selection is transfinited the hidden layer node number of learning machine, training pattern number and neuron emergency function, and removal is transfinited study
Original weight term in machine.
4. the three-dimensional CAD model dividing method according to claim 1 based on the learning machine classifier that transfinites, characterized in that
Described the step of each face of three-dimensional CAD model is classified and marked using the learning machine that transfinites includes:
Several neuron probability values in each face of three-dimensional CAD model are calculated using the learning machine that transfinites；
Several neuron probability values in each face are normalized, the label probability value in the face is obtained；
Classified using the label probability value in each face to each face of three-dimensional CAD model, distinguishes the flat of three-dimensional CAD model
Face, convex surface, concave surface.
5. the three-dimensional CAD model dividing method according to claim 1 based on the learning machine classifier that transfinites, characterized in that
The construction method of the adjacent label figure of the attribute of the three-dimensional CAD model are as follows:
The data structure of the adjacent label figure of defined attribute, including adjacency and concavity and convexity；
It traverses each face of threedimensional model and extracts all properties in each face, the corresponding node of the adjacent label figure of creation attribute；
Identify the syntople between each face of threedimensional model, the side of the adjacent label figure of creation attribute.
6. the three-dimensional CAD model dividing method according to claim 1 based on the learning machine classifier that transfinites, characterized in that
The dividing method of the adjacent label figure of the attribute are as follows:
It is split according to the attribute of the adjacent label figure interior joint of attribute and line label figure G adjacent to attribute, obtains several
Regional area subgraph constitutes regional area collection S；
The line between each node and node in regional area collection S is deleted in the adjacent label figure G of dependence, is obtained new
The adjacent label figure G ' of attribute；
If new attribute linkage flag figure G ' is sky, show to have been completed all nodes and company in the adjacent label figure of attribute
The segmentation of line；
If in new attribute linkage flag figure G ' including the subgraph of Combination node, divide again according to first identification segmentation caviton figure convex
The principle of subgraph divides the attribute linkage flag figure G ' comprising Combination node again, until it is adjacent to obtain new attribute
Label figure is empty.
7. the three-dimensional CAD model dividing method according to claim 1 based on the learning machine classifier that transfinites, characterized in that
The cohesion degree and region couples for further including the steps that the adjacent label figure segmentation of setting regions cohesion degree, attribute are spent, the region
Cohesion degree is the average degree of each node in the adjacent label figure of attribute；The cohesion degree of the attribute linkage flag figure segmentation is three-dimensional
The mean value of the corresponding each regional area cohesion degree of CAD model；The region couples degree is any two office in regional area collection S
Line of the node in the corresponding attribute linkage flag figure of three-dimensional CAD model in the subgraph of portion region.
8. the three-dimensional CAD model dividing method according to claim 1 based on the learning machine classifier that transfinites, characterized in that
The adjacent label figure of attribute after described pair of segmentation merges optimization method are as follows:
According to the cohesion degree expression formula of the adjacent label figure segmentation of attribute, the cohesion degree of the regional area collection S obtained after segmentation is calculated,
New regional area collection S ' is obtained after update；
Each regional area subgraph G in localized region collection S 'i' analyzed, several alternative merging subgraphs are obtained, are constituted
Alternative set A；
It is selected from alternative set A and GiThe maximum subgraph of ' degree of coupling merges processing；
According to the cohesion degree of the adjacent label figure segmentation of attribute, the cohesion degree of regional area collection S ' after segmentation is calculated, before segmentation
The cohesion degree of regional area collection S is compared with the cohesion degree of the regional area collection S ' after segmentation, if regional area collection S's ' is interior
Poly- degree is less than the cohesion degree of regional area collection S, then exports regional area collection S '.
9. the three-dimensional CAD model dividing method according to claim 1 based on the learning machine classifier that transfinites, characterized in that
Each regional area subgraph G in the localized region collection S 'i' the method analyzed are as follows:
By regional area subgraph G each in regional area collection S 'i' attribute analyzed；
Judge whether there is the node of subgraph and the concavity of line and regional area subgraph Gi' consistent；
The node of subgraph and the concavity of connection and regional area subgraph G if it existsi' identical, the then conjunction by the subgraph alternately
And subgraph.
10. a kind of three-dimensional CAD model segmenting device based on the learning machine classifier that transfinites, characterized in that including memory, processing
On a memory and the computer program that can run on a processor, when processor execution described program, is realized for device and storage
Following steps, comprising:
It calculates feature corresponding to each face of three-dimensional CAD model and describes operator；
Feature based on all faces describes operator, and trained and test is transfinited learning machine；
Each face of three-dimensional CAD model is classified and marked using learning machine is transfinited；
Based on classification results, the adjacent label figure of attribute of three-dimensional CAD model is constructed；
The adjacent label figure of attribute is split；
The adjacent label figure of attribute after segmentation is closed using the maximum cohesion degree of the adjacent label figure segmentation of attribute as objective function
And optimize, obtain multiple regional areas.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810758421.9A CN109063271B (en) | 2018-07-11 | 2018-07-11 | Three-dimensional CAD model segmentation method and device based on ultralimit learning machine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810758421.9A CN109063271B (en) | 2018-07-11 | 2018-07-11 | Three-dimensional CAD model segmentation method and device based on ultralimit learning machine |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109063271A true CN109063271A (en) | 2018-12-21 |
| CN109063271B CN109063271B (en) | 2022-12-20 |
Family
ID=64815988
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810758421.9A Active CN109063271B (en) | 2018-07-11 | 2018-07-11 | Three-dimensional CAD model segmentation method and device based on ultralimit learning machine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109063271B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111079208A (en) * | 2019-11-20 | 2020-04-28 | 杭州电子科技大学 | Particle swarm optimization algorithm-based method for identifying surface correspondence between CAD models |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090092299A1 (en) * | 2007-10-03 | 2009-04-09 | Siemens Medical Solutions Usa, Inc. | System and Method for Joint Classification Using Feature Space Cluster Labels |
| JP2013015958A (en) * | 2011-07-01 | 2013-01-24 | Hitachi Ltd | Dependence evaluation device among software development deliverable and evaluation method |
| CN106327506A (en) * | 2016-08-05 | 2017-01-11 | 北京三体高创科技有限公司 | Probability-partition-merging-based three-dimensional model segmentation method |
-
2018
- 2018-07-11 CN CN201810758421.9A patent/CN109063271B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090092299A1 (en) * | 2007-10-03 | 2009-04-09 | Siemens Medical Solutions Usa, Inc. | System and Method for Joint Classification Using Feature Space Cluster Labels |
| JP2013015958A (en) * | 2011-07-01 | 2013-01-24 | Hitachi Ltd | Dependence evaluation device among software development deliverable and evaluation method |
| CN106327506A (en) * | 2016-08-05 | 2017-01-11 | 北京三体高创科技有限公司 | Probability-partition-merging-based three-dimensional model segmentation method |
Non-Patent Citations (2)
| Title |
|---|
| 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》 * |
| 皇甫中民等: "面向三维CAD模型检索的模型分割方法", 《机械科学与技术》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111079208A (en) * | 2019-11-20 | 2020-04-28 | 杭州电子科技大学 | Particle swarm optimization algorithm-based method for identifying surface correspondence between CAD models |
| CN111079208B (en) * | 2019-11-20 | 2024-01-23 | 杭州电子科技大学 | Particle swarm algorithm-based CAD model surface corresponding relation identification method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109063271B (en) | 2022-12-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhang et al. | Solving dynamic traveling salesman problems with deep reinforcement learning | |
| CN105741355B (en) | A kind of block dividing method of triangle grid model | |
| CN109002834B (en) | Fine-grained image classification method based on multi-modal representation | |
| CN101093559B (en) | Method for constructing expert system based on knowledge discovery | |
| CN109801260A (en) | The recognition methods of livestock number and device | |
| CN106803253A (en) | Rock 3-D view crack extract method | |
| CN115438272A (en) | Group discovery system of attribute network | |
| CN110008602B (en) | Road network selection method considering multi-feature coordination under large scale | |
| CN113255895A (en) | Graph neural network representation learning-based structure graph alignment method and multi-graph joint data mining method | |
| CN107944479A (en) | Disease forecasting method for establishing model and device based on semi-supervised learning | |
| CN108595631B (en) | Three-dimensional CAD model double-layer retrieval method based on graph theory | |
| CN111710022A (en) | Rapid forming slice processing method for avoiding contour intersection | |
| CN110830291B (en) | Node classification method of heterogeneous information network based on meta-path | |
| CN105205537B (en) | A kind of device and method of the feature machining Process Knowledge Representation reasoning based on body | |
| CN108683658A (en) | Industry control network Traffic Anomaly recognition methods based on more RBM network structions benchmark models | |
| CN113514072B (en) | Road matching method oriented to navigation data and large-scale drawing data | |
| CN114742564A (en) | False reviewer group detection method fusing complex relationships | |
| CN109063271A (en) | A kind of three-dimensional CAD model dividing method and device based on the learning machine that transfinites | |
| CN110851733A (en) | Community discovery and emotion interpretation method based on network topology and document content | |
| CN108108700B (en) | Pig feature region identification method based on chord axis transformation | |
| CN112085837B (en) | Three-dimensional model classification method based on geometric shape and LSTM neural network | |
| CN112711912A (en) | Air quality monitoring and alarming method, system, device and medium based on cloud computing and machine learning algorithm | |
| Singaravel et al. | Explainable deep convolutional learning for intuitive model development by non–machine learning domain experts | |
| CN112465253B (en) | Method and device for predicting links in urban road network | |
| CN112967333B (en) | Complex point cloud skeleton extraction method and system based on grading |
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 |