A 3D CAD model similar structure retrieval method based on local feature representation

Abstract

The present invention proposes a method for retrieving similar structures of 3D CAD models based on local feature representation. The 3D CAD model is decomposed into multiple local features according to the aggregation characteristics of B-rep elements on the model surface, and the shape characterization vector and neighbor set of the local features are used as their descriptors. On this basis, by calculating the matching local features between two models, the model surface area corresponding to the final matching local features is used as the similar structure, thereby realizing the retrieval of similar structure information. The method for expressing local features of 3D CAD models proposed by the present invention is carried out under B-rep constraints, and is more suitable for solid models commonly used in the field of product design compared with existing methods. In the method for searching for similar structures based on local features in the present invention, the complex structure analysis process is transformed into the matching calculation of several simple local features, thereby reducing the calculation cost.

Description

Technical Field

[0001] The present invention belongs to the field of computer-aided design, and particularly relates to a method for retrieving similar structures of 3D CAD models based on local feature representation. Background Art

[0002] In modern industrial enterprises, the application scope of CAD models has extended from the upstream stage of product structure representation to the downstream stages of process, manufacturing, assembly, and maintenance. Long-term design activities have enabled enterprises to accumulate a large number of CAD models, and the empirical knowledge and design wisdom hidden behind these models have important reusability value for new product development. In actual product design, different products evolve through continuous reference and integration to meet changing functional requirements, resulting in rich design correlations in their local structures. Therefore, the technology of local feature representation and matching of CAD models has become one of the research hotspots in recent years.

[0003] The patent "A method for extracting local spherical harmonic features of a 3D model (CN103700135A, publication date: January 4, 2017)" proposes a method for local feature representation of 3D model shape information. This method performs interpolation resampling on the surface of the 3D model with scale normalization, and then generates a depth image; determines the positions of local centers by grid division, and performs local spherical sampling on each local center using the method of concentric spherical shells; finally, calculates the spherical harmonic feature vectors corresponding to each local center, and obtains the local spherical harmonic feature vector of the entire 3D model using the method of weighted combination. The patent "A method for local matching of 3D models (CN102063719A, publication date: January 16, 2013)" proposes a method for local feature representation and calculation for local matching of 3D models. This method calculates the bending saliency of any vertex of the 3D model, and forms a local vertex set through region growing of the sorted list of curvatures; performs quadratic surface fitting on the local vertex set, and the fitting region is the local sub-block of the 3D model; finally, through the extraction of the shape features of the local sub-block of the 3D model, the feature comparison and local matching of the local sub-block of the 3D model are carried out to achieve the global feature comparison and global matching of the 3D model. The above methods establish a local feature representation method for 3D model shape information from a geometric level, and have good performance in terms of accuracy and efficiency when solving the problem of local matching of 3D models.

[0004] In the local similarity analysis of solid models, there is a dependency relationship between B-rep (Boundary Representation) and local features. However, the above existing technologies do not well consider the dependency relationship between B-rep (Boundary Representation) and local features, and these technologies are difficult to support the local similarity analysis of solid models commonly used in the field of product design. Summary of the Invention

[0005] The technical problem solved by the present invention is: in order to solve the problem that similar structures between three-dimensional CAD models are difficult to locate, the present invention provides a three-dimensional CAD model similar structure retrieval method based on local feature expression, establishes a B-rep feature segmentation and quantitative description process, decomposes the three-dimensional CAD model into multiple local features according to the aggregation characteristics of the B-rep elements on the model surface, and uses the shape representation vector and neighbor set of the local feature as its descriptor. On this basis, by calculating the matching local features between the two models, the model surface area corresponding to the final matching local features is taken as the similar structure, thereby realizing the retrieval of similar structure information.

[0006] The technical solution of the present invention is: a method for retrieving similar structures of three-dimensional CAD models based on local feature expression, comprising the following steps:

[0007] Step 1: Construct a vertex adjacency graph VAG for the 3D CAD model P in B-Rep form;

[0008] Step 2: Using VAG as input, search for local features in the 3D CAD model P;

[0009] Step 3: For each local feature lf in the model P i , a weighted shape distribution algorithm is established to transform each local feature into a feature vector reflecting its shape information;

[0010] Step 4: Determine whether there are the same surface elements between local features;

[0011] Step 5: Combine steps 3 and 4 to establish each local feature lf in the model P i The descriptor lf i =(sv i ,ns i );

[0012] Step 6: For any two local features, a matching judgment method of local features is established from two aspects: shape vector similarity and neighbor vector similarity measurement;

[0013] Step 7: Combine the query model input by the user with the 3D model in the database, and obtain similar local structures between models by comparing local features.

[0014] A further technical solution of the present invention is: in step 1, the expression of VAG is: VAG = (V, E); wherein V = (v1, v1, ..., v n ) is the set of vertices in the graph, each vertex corresponds to a face on the model, and n is the number of faces in the model; E = {<v i ,v j >|h1(v i, v j ) = 1, 1 ≤ i, j ≤ n, i ≠ j} is the set of edges in the graph, h1(·) is a function for judging the topological relationship between faces, h1(v i , v j ) = 1 indicates that there is at least one contact point between the faces corresponding to v i and v j in the model.

[0015] A further technical solution of the present invention is that in the step 2, the local feature is defined as: the geometric region formed by the faces corresponding to all vertices in each maximum clique of the model P in the model P, and this geometric region is used as a local feature of the model P.

[0016] A further technical solution of the present invention is that in the step 2, the local feature of the model P is expressed as:

[0017] P = {lf1, lf2, …, lf n'}

[0018] where lf i is the i-th local feature in the model P, and n' is the number of local features; by judging whether a face belongs to a local feature, the inclusion relationship between the local feature and the model face element is expressed as:

[0019]

[0020] where l i,j = 1 (1 ≤ i ≤ n, 1 ≤ j ≤ n') indicates that the face v i is a constituent face of the local feature lf j .

[0021] A further technical solution of the present invention is that the step 3 includes the following sub-steps:

[0022] Step S3.1: For the local feature containing n lf face elements, the sampling probability p k of the k-th face is:

[0023]

[0024] S k represents the area value of the k-th face; ω represents the scaling ratio;

[0025] Step S3.2: Randomly generate m2 sampling points on the local feature surface to form a surface point set; among them, the ratio of the number of sampling points assigned to each face is p1: p2:...: p nlf ; each sampling point is scaled once according to the following formula:

[0026]

[0027] 'point' represents the coordinate value of the sampling point on the k-th surface; point = (x, y, z) represents the coordinate value of the scaled sampling point; O k represents the centroid point of the k-th surface;

[0028] Step S3.3: Randomly select two sampling points point1 = (x1, y1, z1) and point2 = (x2, y2, z2) and calculate the Euclidean distance between the two points, that is:

[0029]

[0030] Step S3.4: Repeat S3.3 for m2 times;

[0031] Step S3.5: Construct an equidistant histogram with m groups based on the m2 distance values to represent the distribution of the sampling distance values. The value of each column in the histogram is calculated according to the following formula:

[0032]

[0033] where, N u represents the frequency of the distance values falling in the u-th group; h u represents the corresponding frequency;

[0034] Step S3.6: The shape information of the local feature lf i can be represented as an m-dimensional feature vector:

[0035]

[0036] A further technical solution of the present invention is that the value of m2 in the step 3.2 ranges from 1×10 4 to 1×10 5 inclusive.

[0037] A further technical solution of the present invention is to judge whether there are the same surface elements between local features by the following formula:

[0038]

[0039] For any local feature lf i , all local features lf i that satisfy od(lf j ), lf j are used as its neighbor vector ns i , that is:

[0040] ns i = {lf j : lf j ∈P, od(lfi ,lf j ) = 1}.

[0041] A further technical solution of the present invention is that the step 6 includes the following sub-steps:

[0042] Step S6.1: Calculate the similarity of the shape vectors and using the Manhattan distance:

[0043]

[0044] Step S6.2: The similarity calculation method of the neighbor sets ns1 and ns2 is as follows:

[0045] disn(ns1, ns2) = max(mhd(ns1, ns2), mhd(ns2, ns1))

[0046] where mhd(ns1, ns2) and mhd(ns2, ns1) are calculated using the following formulas respectively:

[0047]

[0048] where a and b respectively represent the shape vectors of a local feature in ns1 and ns2, and |ns1| represents the number of local features in ns1;

[0049] Step S6.3: Select the threshold parameters ε1 and ε2 (0 < ε1, ε2 < 1). If disv(sv1, sv2) < ε1 and at the same time

[0050] disn(ns1, ns2) < ε2, then it is considered that lf1 and lf2 are successfully matched.

[0051] A further technical solution of the present invention is that the step 7 includes the following sub-steps:

[0052] Step S7.1: Calculate the similarity of each local feature in model P and the local features in Q according to step 6. The final set of matched local features is:

[0053]

[0054] Step S7.2: According to the mapping relationship matrix between the local features and the face elements of the 3D model, take the face elements corresponding to all the local features in set P' as the region where the similar structure in model P is located under the condition of the query model Q;

[0055] Step S7.3: Search each model in the 3D model library, and output the calculated similar structure region as the result to complete the search for similar structures of the 3D model.

[0056] Advantages of the Invention

[0057] The technical effects of the present invention are as follows: Compared with the prior art, the present invention has the following beneficial effects:

[0058] 1. The method for expressing local features of a 3D CAD model proposed by the present invention is carried out under the B-rep constraint and is more suitable for the solid models commonly used in the product design field compared with the existing methods.

[0059] 2. In the present invention, the method for searching similar structures based on local features transforms the complex structure analysis process into the matching calculations of several simple local features, thereby reducing the calculation cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is the overall flowchart of the local feature extraction and similar structure search of the 3D CAD model in the present invention.

[0061] Figure 2 is an example of the CAD model and its included face elements in the specific implementation manner of the method of the present invention.

[0062] Figure 3 is an example of the VAG of the CAD model in the specific implementation manner of the method of the present invention.

[0063] Figure 4 is an example of the local features and their descriptors of the CAD model in the specific implementation manner of the method of the present invention.

[0064] Figure 5 is the 3D model library for demonstrating the similar structure search in the specific implementation manner of the method of the present invention.

[0065] Figure 6 is an example of the search for similar structures in the specific implementation manner of the method of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0066] The present invention will be described in detail below with reference to specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be noted that those of ordinary skill in the art can make several changes and improvements without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0067] Refer to Figures 1-6 , this method includes the following steps:

[0068] Step S1: For a solid model P in B-Rep form, establish the following vertex adjacency graph (VAG) according to the topological relationship between the face elements in the model:

[0069] VAG = (V, E) (1)

[0070] Wherein, V = (v1, v1, …, v n ) is the set of vertices in the graph, each vertex corresponding to a face on the model, and n is the number of faces in the model; E = {<v i , v j >| h1(v i , v j ) = 1, 1 ≤ i, j ≤ n, i ≠ j} is the set of edges in the graph, and h1() is a function for judging the topological relationship between faces. h1(v i , v j ) = 1 indicates that there is at least one contact point between the faces corresponding to v i and v j in the model.

[0071] Step S2: Using VAG as the input, adopt algorithms such as the BK (Bron-Kerbosch) algorithm, IBK (Improved Bron-Kerbosch) algorithm, and genetic algorithm to search for all existing maximum cliques therein. All the faces corresponding to the vertices in each maximum clique form a geometric region in the model P, and this region is used as a local feature of the model P. The local features of the model P are expressed as:

[0072] P = {lf1, lf2, …, lf n'}} (2)

[0073] Wherein, lf i is the i-th local feature in the model P (each local feature corresponds to a maximum clique obtained by the maximum clique search algorithm), and n' is the number of local features. By judging whether a face belongs to a local feature, the inclusion relationship between the local feature and the model face elements is expressed as:

[0074]

[0075] Wherein, l i,j = 1 (1 ≤ i ≤ n, 1 ≤ j ≤ n') indicates that the face v i is a constituent face of the local feature lf j .

[0076] Step S3: For each local feature lf i in the model P, establish a weighted shape distribution algorithm to convert each local feature into a feature vector reflecting its shape information, which specifically includes the following steps:

[0077] Step S3.1: For a local feature containing n lf face elements, the sampling probability p of the k-th facek is:

[0078]

[0079] S k represents the area value of the k-th face; ω represents the scaling ratio.

[0080] Step S3.2: Randomly generate m2 sampling points on the local feature surface to form a surface point set. Among them, the ratio of the number of sampling points assigned to each face is p1:p2:...:p nlf . Considering both sampling accuracy and computational efficiency, m2 is generally in the range of 1×10 4 ~1×10 5 . For the convenience of shape difference recognition, each sampling point is scaled once according to Equation (5):

[0081]

[0082] point' represents the coordinate value of the sampling point on the k-th face; point = (x, y, z) represents the coordinate value of the scaled sampling point; O k represents the centroid point of the k-th face.

[0083] Step S3.3: Randomly select two sampling points point1 = (x1, y1, z1) and point2 = (x2, y2, z2) and calculate the Euclidean distance between the two points, that is:

[0084]

[0085] Step S3.4: Repeat S3.3 m2 times;

[0086] Step S3.5: Construct an equal-interval histogram with m groups based on the m2 distance values to represent the distribution of the sampling distance values. The value of each column in the histogram is calculated according to Equation (7):

[0087]

[0088] where N u represents the frequency of the distance values falling into the u-th group; h u represents the corresponding frequency.

[0089] Step S3.6: The shape information of the local feature lf i can be represented as an m-dimensional feature vector:

[0090]

[0091] Step S4: Determine whether there are the same face elements between local features through Equation (9):

[0092]

[0093] For any local feature lf i , all local features lf that satisfy od(lf i , lf j ) = 1 are taken as its neighbor vector ns j , that is: i

[0094] ns i = {lf j : lf j ∈ P, od(lf i , lf j ) = 1} (10)

[0095] Step S5: Combining steps S3 and S4, the descriptor of each local feature lf in model P can be expressed as: i

[0096] lf i = (sv i , ns i ) (11)

[0097] Step S6: For any two local features lf1 = (sv1, ns1) and lf2 = (sv2, ns2), a matching judgment method for local features is established from two aspects: the similarity of shape vectors and the similarity measure of neighbor vectors, specifically including the following steps:

[0098] Step S6.1: Calculate the similarity of shape vectors using the Manhattan distance and :

[0099]

[0100] Step S6.2: The similarity calculation method for neighbor sets ns1 and ns2 is as follows:

[0101] disn(ns1, ns2) = max(mhd(ns1, ns2), mhd(ns2, ns1)) (13)

[0102] where mhd(ns1, ns2) and mhd(ns2, ns1) are calculated using Equation (14) respectively:

[0103]

[0104] where a and b represent the shape vectors of a local feature in ns1 and ns2 respectively, and |ns1| represents the number of local features in ns1.

[0105] Step S6.3: Select threshold parameters ε1 and ε2 (0 < ε1, ε2 < 1). If disv(sv1, sv2) < ε1 and

[0106] disn(ns1, ns2) < ε2, then it is considered that lf1 and lf2 are successfully matched.

[0107] Step S7: Assume that the query model input by the user is A 3D model in the database is Obtain the similar local structures between the models by comparing the local features. The specific steps are as follows:

[0108] Step S7.1: Calculate the similarity between each local feature in model P and the local features in Q according to Step 6. The final set of matched local features is:

[0109]

[0110] Step S7.2: According to the mapping relationship matrix between the local features and the 3D model surface elements in Equation (3), use the surface elements corresponding to all the local features in set P' as the regions where the similar structures in model P are located under the condition of the query model Q.

[0111] Step S7.3: For each model in the 3D model library, repeat Step S7.3. Output the calculated similar structure regions as the results to complete the search for similar structures of the 3D models.

[0112] Refer to Figure 1 , Figure 1 which is a schematic diagram of the steps of a method for retrieving similar structures of 3D CAD models based on local features provided by an embodiment of the present invention, and it includes the following steps:

[0113] Step S1: According to the composition of the 3D CAD model surface elements as shown in Figure 2 If there are the same vertices between any two faces, then construct a connecting edge in the corresponding topological space to form a VAG containing 9 vertices as shown in Figure 3 :

[0114] VAG = (V, E) (16)

[0115] Among them, V = (v1, v1,..., v9) is the vertex set of VAG, and each vertex corresponds to a face on the model; E = {<v i , v j >|h1(v i , v j ) = 1, 1 ≤ i, j ≤ 9, i ≠ j} is the edge set of VAG, and h1(·) is a function for judging the topological relationship between faces, h1(v i , vj ) = 1 indicates that there is at least one contact point corresponding to v in the model i and v j corresponds to at least one contact point.

[0116] Step S2: Using VAG in Figure 3 as the input, the BK algorithm is used to search for all maximal cliques therein, and 6 maximal cliques and their corresponding 6 local features are obtained as Figure 4 lf1 to lf6 in

[0117] P = {lf1, lf2,..., lf6} (17)

[0118] where lf i is the i-th local feature (maximal clique) in model P, and n' is the number of local features. The relationship between each local feature and the model face elements is expressed as:

[0119]

[0120] where l i,j = 1 (1 ≤ i ≤ n, 1 ≤ j ≤ n') indicates that face v i is a constituent face of the local feature lf j .

[0121] Step S3: For each local feature lf i in model P (1 ≤ i ≤ 6), a weighted shape distribution algorithm is established to convert each local feature into a feature vector reflecting its shape information, which specifically includes the following steps:

[0122] Step S3.1: For a local feature containing n lf face elements, the sampling probability p k of each face is:

[0123]

[0124] S k represents the area value of the k-th face; ω represents the scaling ratio.

[0125] Step S3.2: Considering the sampling accuracy and computational efficiency comprehensively, 10,000 sampling points are randomly generated on the local feature surface to form a surface point set. Among them, the ratio of the number of sampling points assigned to each face is For the convenience of shape difference recognition, each sampling point is scaled once:

[0126]

[0127] point' represents the coordinate value of the sampling point on the k-th face; point = (x, y, z) represents the coordinate value of the scaled sampling point; Ok Represents the centroid point of the k-th face.

[0128] Step S3.3: Randomly select two sampling points point1 = (x1, y1, z1) and point2 = (x2, y2, z2) and calculate the Euclidean distance between the two points, that is:

[0129]

[0130] Step S3.4: Repeat step S3.3 100,000 times;

[0131] Step S3.5: Construct an equidistant histogram with 50 groups based on 100,000 distance values to represent the distribution of the sampling distance values. The value of each column in the histogram is calculated according to Equation (3):

[0132]

[0133] where N u Represents the frequency of the distance values falling in the u-th group; h u Represents the corresponding frequency.

[0134] Step S3.6: The shape information of the local feature lf i Can be represented as a 50-dimensional feature vector The shape vectors sv1 to sv6 of 6 local features are shown in Figure 4 .

[0135] Step S4: Determine whether there are the same face elements between local features through Equation (7):

[0136]

[0137] For any local feature lf i , all local features lf that satisfy od(lf i , lf j ) = 1 are used as its neighbor vector ns j . The neighbor vectors ns1 to ns6 of 6 local features are shown in i . Figure 4 .

[0138] Step S5: Combining steps S3 and S4, the descriptor of each local feature lf in model P i Can be expressed as:

[0139] lf i = (sv i , ns i ) (24)

[0140] Step S6: For any two local features lf1 = (sv1, ns1) and lf2 = (sv2, ns2), perform matching judgment of local features from two aspects: shape vector similarity and neighbor vector similarity measure.

[0141] Step S6.1: Calculate the similarity of shape vectors using the Manhattan distance and of:

[0142]

[0143] Step S6.2: The similarity calculation method of neighbor sets ns1 and ns2 is as follows:

[0144] disn(ns1, ns2) = max(mhd(ns1, ns2), mhd(ns2, ns1)) (26)

[0145] where mhd(ns1, ns2) and mhd(ns2, ns1) are calculated using Equation (12) respectively:

[0146]

[0147] where a and b represent the shape vectors of a local feature in ns1 and ns2 respectively, and |ns1| represents the number of local features in ns1.

[0148] Step S6.3: Select threshold parameters ε1 = 0.1 and ε2 = 0.2. If disv(sv1, sv2) < ε1 and at the same time

[0149] disn(ns1, ns2) < ε2, it is considered that lf1 and lf2 are successfully matched.

[0150] Step S7: Construct a three-dimensional model library as shown in Figure 5 A three-dimensional model in the model library is For the query model input by the user Obtain the similar local structures between models by comparing local features. The specific steps are as follows:

[0151] Step S7.1: Calculate the similarity between each local feature in model P and the local features in Q according to Step 6. The finally matched local feature set is:

[0152]

[0153] Step S7.2: According to the mapping relationship matrix between local features and three-dimensional model face elements in Equation (3), take the face elements corresponding to all local features in set P' as the regions where the similar structures in model P are located under the condition of the query model Q.

[0154] Step S7.3: For each model in the 3D model library, repeat Step S7.3. Output the calculated similar structure region as the result to complete the search for similar structures of 3D models. Figure 6 An example of local structure search between models is shown.

[0155] It can be seen from this that the method for searching similar structures of 3D CAD models according to the present invention converts complex structure information into several local features with discrimination ability, and realizes the retrieval of similar structures of models through the shape and neighbor similarity measurement between local features, which can provide technical support for the realization of related services such as information reuse in the product design process and 3D model library management.

Claims

1. A 3D CAD model similar structure retrieval method based on local feature expression, characterized in that It includes the following steps: Step 1: For the 3D CAD model P in B-Rep form, construct a vertex adjacency graph VAG; Step 2: Using VAG as the input, search for local features in the 3D CAD model P, where the local feature is defined as: the geometric region formed by the faces corresponding to all vertices in each maximum clique of the model P in the model P, and this geometric region is used as a local feature of the model P; Step 3: For each local feature lf in model P i , establish a weighted shape distribution algorithm to transform each local feature into a feature vector reflecting its shape information; It includes the following sub-steps: Step S3.1: For the local feature containing n lf face elements, the sampling probability p of the k-th face k is as follows: S k represents the area value of the k-th face; ω represents the scaling ratio; Step S3.2: Randomly generate m2 sampling points on the local feature surface to form a surface point set; among them, the ratio of the number of sampling points assigned to each face is Each sampling point is scaled once according to the following formula: point' represents the coordinate value of the sampling point on the k-th surface; point = (x, y, z) represents the coordinate value of the scaled sampling point; O k represents the centroid point of the k-th surface; n is the number of surfaces in the model; Step S3.3: Randomly select two sampling points point1 = (x1, y1, z1) and point2 = (x2, y2, z2) and calculate the Euclidean distance between the two points, that is: Step S3.4: Repeat S3.3 for m2 times; Step S3.5: Construct an equidistant histogram with m groups based on the m2 distance values to represent the distribution of the sampled distance values. The value of each column in the histogram is calculated according to the following formula: Among them, N u represents the frequency of the distance values falling in the u-th group; h u represents the corresponding frequency; Step S3.6: The shape information of the local feature lf i can be represented as an m-dimensional feature vector: Step 4: Determine whether there are the same face elements between local features; Step 5: Combine Step 3 and Step 4 to establish the descriptor lf of each local feature lf in model P i of lf i = (sv i , ns i ); ns i is the neighbor vector of the local feature lf j : Step 6: For any two local features, establish a matching judgment method for local features from two aspects: shape vector similarity and neighbor vector similarity measure; Step 7: Combine the query model input by the user with the 3D models in the database, and obtain the similar local structures between the models through the comparison of local features.

2. The 3D CAD model similar structure retrieval method based on local feature expression according to claim 1, characterized in that In the said step 1, the VAG expression is: VAG = (V, E); where V = (v1, v1, …, v n ) is the set of vertices in the graph, each vertex corresponding to a face on the model, and n is the number of faces in the model; E = {<v i , v j >|h1(v i , v j ) = 1, 1 ≤ i, j ≤ n, i ≠ j} is the set of edges in the graph, h1() is a function for judging the topological relationship between faces, and h1(v i , v j ) = 1 means that there is at least one contact point between the faces corresponding to v i and v j in the model.

3. The 3D CAD model similar structure retrieval method based on local feature expression according to claim 1, characterized in that, In the said Step 2, the local feature of the model P is expressed as: P = {lf1, lf2, …, lf n'} Among them, lf i is the i-th local feature in model P, and n' is the number of local features; by determining whether a face belongs to a local feature, the inclusion relationship between the local feature and the model face element is expressed as: where l i,j = 1 (1 ≤ i ≤ n, 1 ≤ j ≤ n') indicates that the face v i is a constituent face of the local feature lf j .

4. The 3D CAD model similar structure retrieval method based on local feature expression according to claim 1, characterized in that The value of m2 in step 3.2 ranges from 1×10 4 to 1×10 5 .

5. A 3D CAD model similar structure retrieval method based on local feature expression according to claim 1, characterized in that, Judge whether there are the same face elements between local features through the following formula: For any local feature lf i , all local features lf i that satisfy od(lf j , lf j ) = 1 are used as its neighbor vector ns i , that is: ns i = {lf j :lf j ∈ P, od(lf i , lf j ) = 1}.

6. The 3D CAD model similar structure retrieval method based on local feature expression according to claim 1, wherein The said Step 6 includes the following sub-steps: Step S6.1: Calculate the similarity between the shape vectors using the Manhattan distance and : Step S6.2: The similarity calculation method for the neighbor sets ns1 and ns2 is as follows: disn(ns1, ns2) = max(mhd(ns1, ns2), mhd(ns2, ns1)) where mhd(ns1, ns2) and mhd(ns2, ns1) are calculated respectively using the following formula: where a and b respectively represent the shape vectors of a local feature in ns1 and ns2, and |ns1| represents the number of local features in ns1; Step S6.3: Select threshold parameters ε1 and ε2, where 0 < ε1, ε2 < 1. If disv(sv1, sv2) < ε1 and at the same time disn(ns1, ns2) < ε2, then it is considered that lf1 and lf2 are successfully matched.

7. The 3D CAD model similar structure retrieval method based on local feature expression according to claim 6, characterized in that, The said Step 7 includes the following sub-steps: Step S7.1: Calculate the similarity between each local feature in the model P and the local features in Q according to Step 6. The final set of matched local features is: Step S7.2: According to the mapping relationship matrix between local features and 3D model face elements, take the face elements corresponding to all local features in the set P' as the regions where the similar structures in the model P are located under the condition of the query model Q; Step S7.3: For each model in the 3D model library, perform a search, and output the calculated similar structure region as the result to complete the search for similar structures of 3D models.

Images