BIM object recognition and semantic data automatic generation method based on deep learning

Abstract

The invention specifically provides a BIM object recognition and semantic data automatic generation method based on deep learning. The method comprises the steps of obtaining a BIM data set; preprocessing the BIM data set to obtain a preprocessed BIM data set; the preprocessed BIM data set is substituted into the trained CDBN model, so that the multiple pieces of BIM data in the OFF formats in the preprocessed BIM data set are recognized and classified, and the classification result of each piece of BIM data in the OFF formats in the preprocessed BIM data set is obtained; and based on the classification result of the BIM data in each OFF format in the preprocessed BIM data set, selectively performing semantic data enhancement on the BIM data in each OFF format in the preprocessed BIM data set to obtain a BIM data set after semantic data enhancement. The reuse efficiency of the BIM data in the building operation stage is improved, and the data creation cost is reduced.

Description

Technical Field

[0001] The present invention relates to the field of BIM technology, and in particular to the field of 3D BIM technology, and specifically provides a method for BIM object recognition and automatic generation of semantic data based on deep learning. Background Art

[0002] BIM technology was introduced in the late 1980s and became widespread in the construction industry in the early 21st century, leveraging authoring tools. It has become a crucial productivity tool. It integrates building-related data, visualizes it through 3D models, and expresses it semantically. However, throughout the lifecycle of a construction project, frequent interdisciplinary data exchange leads to difficulties in migrating existing data, resulting in poor quality and difficulty in reusing it. This semantic data issue directly hinders the full value of BIM. Furthermore, semantic data is difficult to automatically mine, making it difficult to ensure data consistency and accuracy, and resulting in high processing costs.

[0003] Accordingly, this field needs a new BIM object recognition and semantic data automatic generation solution based on deep learning to solve the above problems. Summary of the Invention

[0004] In order to overcome the above-mentioned defects, the present invention is proposed to provide a method for BIM object recognition and automatic generation of semantic data based on deep learning to solve or at least partially solve the technical problems in the prior art of BIM technology in the whole life cycle of construction projects, such as frequent cross-disciplinary data exchange, which leads to difficulty in migrating existing data, poor quality, and difficulty in reuse. The semantic data problem directly restricts the value of BIM, and semantic data is difficult to use for automatic mining, which makes it difficult to ensure data consistency and accuracy, and leads to high processing costs.

[0005] In a first aspect, the present invention provides a method for BIM object recognition and automatic semantic data generation based on deep learning, the method comprising: Acquire a BIM dataset, wherein the BIM dataset includes at least a plurality of BIM data in an IFC format; Preprocessing the BIM dataset to obtain a preprocessed BIM dataset, wherein the preprocessed BIM dataset at least includes a plurality of BIM data in an OFF format and an object category corresponding to each BIM data in the OFF format; Substituting the preprocessed BIM dataset into the trained CDBN model, the multiple BIM data in the OFF format in the preprocessed BIM dataset are identified and classified, and the classification results of each BIM data in the OFF format in the preprocessed BIM dataset are obtained; Furthermore, based on the classification result of each OFF format BIM data in the preprocessed BIM dataset, semantic data enhancement is selectively performed on each OFF format BIM data in the preprocessed BIM dataset to obtain a BIM dataset with enhanced semantic data.

[0006] In one technical solution of the above-mentioned method for BIM object recognition and automatic semantic data generation based on deep learning, obtaining a BIM dataset includes: Acquire multiple BIM data from different data sources; The IFC objects in each BIM data are extracted to form multiple BIM data in IFC format to obtain a BIM dataset.

[0007] In one technical solution of the above-mentioned method for BIM object recognition and automatic semantic data generation based on deep learning, the preprocessing of the BIM dataset to obtain the preprocessed BIM dataset includes: Perform data format conversion on multiple BIM data in IFC format in the BIM dataset to obtain the OFF file corresponding to each BIM data in IFC format; Clean the OFF file corresponding to each BIM data in IFC format to obtain the cleaned OFF file; Based on the IFC categories of multiple BIM data in the IFC format in the BIM dataset, the cleaned OFF files are grouped into categories to obtain the object category corresponding to each OFF format BIM data; Based on the cleaned OFF files and the object categories corresponding to each OFF format BIM data, a preprocessed BIM dataset is constructed.

[0008] In one technical solution of the above-mentioned method for BIM object recognition and automatic semantic data generation based on deep learning, before substituting the preprocessed BIM dataset into the trained CDBN model, the method further trains the CDBN model through the following steps: Build the CDBN model; Create two sets of training data sets based on the two preset basic training data sets; The CDBN model is trained based on two sets of training data sets, and two CDBN models are obtained respectively.

[0009] In one technical solution of the above-mentioned deep learning-based BIM object recognition and semantic data automatic generation method, the construction of the CDBN model includes: Construct multiple CRBM network architectures and stack them to obtain a CDBN model, where each CRBM network architecture includes a visible layer, a hidden layer, and a pooling layer; The CDBN model is expressed by the following formula:

[0010] Where k represents the number of CRBM network architectures, represents the k-th CRBM network architecture, represents the visible layer input data dimension of the k-th CRBM, represents the number of hidden layer neurons of the k-th CRBM, Represents the pooling kernel size of the pooling layer of the k-th CRBM.

[0011] In one technical solution of the above-mentioned method for BIM object recognition and automatic semantic data generation based on deep learning, the creation of two sets of training data sets based on two preset sets of basic training data sets includes: Obtain two preset training data sets, wherein both training data sets include a plurality of CAD object data and an IFC preset category corresponding to each CAD object data; Based on the two preset training data sets, two training data sets are created, wherein both training data sets include a plurality of BIM data in OFF format and an object category corresponding to each BIM data in OFF format.

[0012] In one technical solution of the above-mentioned method for BIM object recognition and automatic semantic data generation based on deep learning, the CDBN model is trained based on two sets of training data sets to obtain two CDBN models respectively, including: Substitute any set of training data sets into the constructed CDBN model, perform convolution operations on the training data sets layer by layer to extract features, and obtain the convolved feature data corresponding to the training data sets; Determining, based on the convolved feature data, a recognition result corresponding to each CAD object in the training data set and a classification result corresponding to each CAD object; Based on the recognition result corresponding to each CAD object in the training data set and the classification result corresponding to each CAD object, selectively updating the model parameters of the CDBN model to obtain an updated CDBN model corresponding to the training data set; Re-execute the steps of "substituting the training dataset into the constructed CDBN model, performing a convolution operation on the training dataset layer by layer to extract features, and obtaining convolved feature data corresponding to the training dataset" and subsequent steps to obtain two CDBN models corresponding to the two sets of training datasets, respectively.

[0013] In one technical solution of the above-mentioned deep learning-based BIM object recognition and semantic data automatic generation method, the method selectively updates the model parameters of the CDBN model based on the recognition result corresponding to each CAD object in the training dataset and the classification result corresponding to each CAD object, thereby obtaining the updated CDBN model corresponding to the training dataset, including: Based on the recognition result corresponding to each CAD object in the training data set and the classification result corresponding to each CAD object, obtaining an optimization combination index of the training data set for the CDBN model; The optimization combination index of the training data set for the CDBN model is obtained by the following formula: ; Among them, L represents the optimization combination index, N represents the number of samples in the training data set, and C represents the number of categories in the IFC preset categories. Represents the actual label of the i-th sample and whether the sample belongs to the c-th category: If it does, then ;otherwise, , represents the probability that the i-th sample belongs to the c-th category predicted by the model; Furthermore, based on the optimized combination index, selectively updating the learning rate and the number of iterations in the model parameters of the CDBN model to obtain an updated CDBN model corresponding to the training data set; Among them, the learning rate in the model parameters of the CDBN model is updated by the following formula: ; in, represents the updated learning rate, represents the learning rate before updating, Represents the preset learning rate attenuation factor, which is taken here ; In addition, the judgment on whether the learning rate in the model parameters of the CDBN model is updated satisfies the following formula: ; in, Represents the current optimized combination index, Represents the last optimized combination index, Represents the preset difference threshold; and, Get the preset fluctuation threshold; In M consecutive iterations, the average value and standard deviation of the optimized combination index of the Mth iteration are obtained; Based on the average and standard deviation of the optimized combination index of the Mth iteration and the preset fluctuation threshold, determine whether to continue iterating the CDBN model; The mean and standard deviation of the optimized combination index of the Mth iteration are obtained by the following formula: ; ; in, Represents the average value of the optimized combination index, represents the standard deviation of the optimized combination index, Represents the optimized combination index of the Mth iteration; The preset fluctuation threshold is obtained by the following formula: ; in, Represents the preset fluctuation threshold, Represents the preset adjustment factor, which is 0.1 here; Based on the average value and standard deviation of the optimized combination index of the Mth iteration and the preset fluctuation threshold, determining whether to continue iterating the CDBN model includes: If the optimization combination index of M consecutive iterations is Fluctuates within a certain range, and ,in, If the difference threshold is exceeded, it is determined that the optimization combination index has no significant downward trend, and the training is stopped, and the number of iterations in the model parameters of the CDBN model is updated with the current number of iterations.

[0014] In one technical solution of the above-mentioned deep learning-based BIM object recognition and semantic data automatic generation method, the preprocessed BIM dataset is substituted into the trained CDBN model to identify and classify multiple BIM data in the OFF format in the preprocessed BIM dataset. The classification results of each OFF format BIM data in the preprocessed BIM dataset include: Obtaining BIM object categories in a preset BIM object database; Based on the preprocessed BIM dataset, determining the trained CDBN model corresponding to the preprocessed BIM dataset; Substitute the preprocessed BIM dataset into the trained CDBN model corresponding to the preprocessed BIM dataset, perform feature extraction on multiple BIM data in OFF format in the preprocessed BIM dataset, and obtain feature data of each BIM data; Determining, based on the characteristic data of each BIM data, a category among preset BIM object categories to which each BIM data belongs; The category in the preset BIM object category to which each BIM data belongs is used as the classification result of each OFF format BIM data in the preprocessed BIM dataset.

[0015] In one technical solution of the above-mentioned method for BIM object recognition and automatic semantic data generation based on deep learning, the preset BIM object database stores multiple BIM object categories and different types of semantic data matching each BIM object category, and the selectively performing semantic data enhancement on each OFF format BIM data in the preprocessed BIM dataset based on the classification result of each OFF format BIM data in the preprocessed BIM dataset includes: If semantic data is missing from any of the OFF format BIM data in the pre-processed BIM dataset, it is determined that semantic data enhancement needs to be performed on the OFF format BIM data; generating, based on a preset BIM object database and the BIM data in the OFF format, at least one type of semantic data that matches the BIM data in the OFF format; At least one type of semantic data matching the BIM data in the OFF format is added to the BIM data in the OFF format to achieve semantic enhancement of the BIM data.

[0016] The above one or more technical solutions of the present invention have at least one or more of the following beneficial effects: By preprocessing the acquired BIM data set to obtain a preprocessed BIM data set, the BIM data is cleaned and the format is unified to improve the accuracy and efficiency of data processing by the CDBN model. By substituting the preprocessed BIM data set into the trained CDBN model, the classification result of each BIM data in the preprocessed BIM data set is obtained, and accurate identification and classification of each BIM data is achieved, thereby providing a unified and reliable foundation for cross-system data interaction and supporting efficient collaboration between different systems in the supply chain. Then, based on the classification result of each BIM data, semantic data enhancement of the BIM data is selectively performed, which solves the data reuse problem caused by inconsistent semantic data, improves the reuse efficiency of BIM data in the building operation stage, reduces the data creation cost, and provides a more accurate and efficient technical means for the digital management of the construction industry, which helps to improve the management level of the entire life cycle of the construction project and avoids the BIM technology in the existing technology. In the entire life cycle of the construction project, due to the frequent cross-disciplinary exchange of data, the existing data is difficult to migrate, the quality is poor, and it is difficult to reuse. The semantic data problem directly restricts BIM Moreover, semantic data is difficult to use for automatic mining, which makes it difficult to ensure data consistency and accuracy, leading to technical problems such as high processing costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The disclosure of the present invention will be more easily understood with reference to the accompanying drawings. Those skilled in the art will readily appreciate that these drawings are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. Furthermore, similar numbers in the drawings represent similar components, wherein: Figure 1 1 is a flowchart of the main steps of a method for BIM object recognition and automatic generation of semantic data based on deep learning according to an embodiment of the present invention; Figure 2 1 is a schematic diagram of a data processing flow of a method for BIM object recognition and automatic semantic data generation based on deep learning according to an embodiment of the present invention; Figure 3 1 is a schematic diagram of a BIM object recognition architecture of a method for BIM object recognition and automatic semantic data generation based on deep learning according to an embodiment of the present invention; Figure 4 This is a CDBN model architecture diagram of a deep learning-based BIM object recognition and semantic data automatic generation method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] Some embodiments of the present invention are described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0019] In the description of the present invention, "module" and "processor" may include hardware, software, or a combination of both. A module may include hardware circuitry, various suitable sensors, communication ports, and memory. It may also include software components, such as program code, or a combination of software and hardware. A processor may be a central processing unit, a microprocessor, an image processor, a digital signal processor, or any other suitable processor. A processor has data and / or signal processing capabilities. A processor may be implemented in software, hardware, or a combination of both. Non-transitory computer-readable storage media include any suitable medium capable of storing program code, such as magnetic disks, hard disks, optical disks, flash memory, read-only memory, random access memory, and the like. The term "A and / or B" refers to all possible combinations of A and B, such as only A, only B, or both A and B. The terms "at least one of A or B" or "at least one of A and B" have similar meanings to "A and / or B" and may include only A, only B, or both A and B. The singular forms "one" and "the" may also include the plural forms.

[0020] Here we first explain some terms involved in the present invention.

[0021] BIM, Building Information Modeling, building information model, three-dimensional model of construction engineering; CDBN model, Convolution Depth Belief Network, convolution deep belief network model; OFF format, Object File Format, represents the geometric structure of a model by describing the polygons on the surface of the object; IFC format, Industry Foundation Classes, a universal data exchange format used in architecture, engineering, and manufacturing; ModelNet10 dataset, a 3D model dataset containing 10 different categories of 3D models, each category containing more than 400 models; ModelNet40 dataset, a 3D shape dataset containing 40 different categories of 3D models, each category containing more than 400 models; CAD object data refers to various information about graphic objects stored in CAD files. This information includes object type, spatial location, layer, color, line type, line width, area, perimeter, etc.

[0022] National BIM Library, NBL, National Building Library.

[0023] See attached Figure 1 , Figure 1 FIG1 is a flow chart of the main steps of a method for BIM object recognition and automatic generation of semantic data based on deep learning according to an embodiment of the present invention. Figures 1 to 4 As shown, the method for BIM object recognition and automatic generation of semantic data based on deep learning in an embodiment of the present invention mainly includes the following steps S101 to S104.

[0024] Step S101: Acquire a BIM dataset, wherein the BIM dataset at least includes a plurality of BIM data in IFC format; Specifically, obtaining the BIM dataset includes: Acquire multiple BIM data from different data sources; The IFC objects in each BIM data are extracted to form multiple BIM data in IFC format to obtain a BIM dataset.

[0025] Step S102: preprocessing the BIM dataset to obtain a preprocessed BIM dataset, wherein the preprocessed BIM dataset at least includes a plurality of BIM data in OFF format and an object category corresponding to each BIM data in OFF format; Specifically, the preprocessing of the BIM dataset to obtain the preprocessed BIM dataset includes: Perform data format conversion on multiple BIM data in IFC format in the BIM dataset to obtain the OFF file corresponding to each BIM data in IFC format; Clean the OFF file corresponding to each BIM data in IFC format to obtain the cleaned OFF file; Based on the IFC categories of multiple BIM data in the IFC format in the BIM dataset, the cleaned OFF files are grouped into categories to obtain the object category corresponding to each OFF format BIM data; Based on the cleaned OFF file and the object category corresponding to each OFF format BIM data, a preprocessed BIM dataset is constructed, wherein the number of multiple OFF format BIM data in the preprocessed BIM dataset is The number of multiple BIM data in IFC format in the BIM dataset.

[0026] Specifically, taking the BIM model of a large commercial building as an example, the model contains many complex building components, such as different types of columns, beams, and floor slabs. Each component is composed of multiple IFC products. Obtaining the BIM dataset includes the following steps: Obtain data from the BIM model; Extracting different types of IFC objects from the BIM model data, wherein each IFC object includes at least a plurality of IFC product data; The extracted IFC objects are used as multiple BIM data in IFC format to obtain a BIM dataset.

[0027] In addition, the BIM dataset is preprocessed to obtain the following preprocessed BIM dataset: The data format of multiple BIM data in the IFC format in the BIM data set is converted by a preset compiler to obtain the OFF file corresponding to each BIM data in the IFC format; Clean up the OFF files corresponding to each BIM data in IFC format, remove abnormal data (such as incorrect geometric patches, incomplete model information, etc.) generated by model conversion, and obtain the cleaned OFF files; Based on different types of IFC objects, the cleaned OFF files are grouped and stored to facilitate detailed classification of the pre-processed BIM dataset.

[0028] Specifically, taking one of the IFC objects as a complex combined air conditioning unit IFC object, and the IFC object containing n IFC products as an example, the BIM dataset is preprocessed to obtain the following preprocessed BIM dataset: Convert the data formats of n IFC products through a preset compiler to obtain n converted OFF files; Perform data cleaning on the converted n OFF files, remove elements irrelevant to the target object, and obtain m cleaned OFF files, and satisfy , that is, get m OFF files related to the core components of the unit; Based on different types of IFC objects, the cleaned OFF files are grouped and stored. The object category corresponding to each OFF file is , then the category set of OFF files is , that is, m OFF files related to the core components of the unit are classified into the air conditioning unit category folder.

[0029] Step S103: Substituting the preprocessed BIM dataset into the trained CDBN model, so as to identify and classify the multiple BIM data in the OFF format in the preprocessed BIM dataset, and obtaining a classification result for each BIM data in the OFF format in the preprocessed BIM dataset; Specifically, before substituting the pre-processed BIM dataset into the trained CDBN model, the method further trains the CDBN model through the following steps: Build the CDBN model; Create two sets of training data sets based on the two preset basic training data sets; The CDBN model is trained based on two sets of training data sets, and two CDBN models are obtained respectively.

[0030] Specifically, the construction of the CDBN model includes: Construct multiple CRBM network architectures and stack them to obtain a CDBN model, where each CRBM network architecture includes a visible layer, a hidden layer, and a pooling layer; The CDBN model can be expressed by the following formula:

[0031] Where k represents the number of CRBM network architectures, represents the k-th CRBM network architecture, represents the visible layer input data dimension of the k-th CRBM, represents the number of hidden layer neurons of the k-th CRBM, Represents the pooling kernel size of the pooling layer of the k-th CRBM.

[0032] Specifically, the creating two sets of training data sets based on the two preset basic training data sets includes: Obtain two preset training data sets, wherein the two training data sets are a ModelNet10 data set and a ModelNet40 data set, and both training data sets include a plurality of CAD object data and an IFC preset category corresponding to each CAD object data; Based on the two preset training data sets, two training data sets are created, wherein both training data sets include a plurality of BIM data in OFF format and an object category corresponding to each BIM data in OFF format.

[0033] In the above embodiment, the ModelNet10 and ModelNet40 datasets are selected as basic training data. These two datasets contain a large number of classified CAD object instances and have categories corresponding to IFC predefined entity types, providing rich samples for training the CDBN model.

[0034] Specifically, the CDBN model is trained based on two sets of training data sets to obtain two CDBN models respectively, including: Substitute any set of training data sets into the constructed CDBN model, perform convolution operations on the training data sets layer by layer to extract features, and obtain the convolved feature data corresponding to the training data sets; Determining, based on the convolved feature data, a recognition result corresponding to each CAD object in the training data set and a classification result corresponding to each CAD object; Based on the recognition result corresponding to each CAD object in the training data set and the classification result corresponding to each CAD object, selectively updating the model parameters of the CDBN model to obtain an updated CDBN model corresponding to the training data set; Re-execute the steps of "substituting the training dataset into the constructed CDBN model, performing a convolution operation on the training dataset layer by layer to extract features, and obtaining convolved feature data corresponding to the training dataset" and subsequent steps to obtain two CDBN models corresponding to the two sets of training datasets, respectively.

[0035] Specifically, selectively updating the model parameters of the CDBN model based on the recognition result corresponding to each CAD object in the training data set and the classification result corresponding to each CAD object to obtain the updated CDBN model corresponding to the training data set includes: Based on the recognition result corresponding to each CAD object in the training data set and the classification result corresponding to each CAD object, obtaining an optimization combination index of the training data set for the CDBN model; The optimization combination index of the training data set for the CDBN model is obtained by the following formula: ; Among them, L represents the optimization combination index, N represents the number of samples in the training data set, and C represents the number of categories in the IFC preset categories. Represents the actual label of the i-th sample and whether the sample belongs to the c-th category: If it does, then ;otherwise, , represents the probability that the i-th sample belongs to the c-th category predicted by the model; Furthermore, based on the optimized combination index, selectively updating the learning rate and the number of iterations in the model parameters of the CDBN model to obtain an updated CDBN model corresponding to the training data set; Among them, the learning rate in the model parameters of the CDBN model is updated by the following formula: ; in, represents the updated learning rate, represents the learning rate before updating, Represents the preset learning rate attenuation factor, which is taken here ; In addition, the judgment on whether the learning rate in the model parameters of the CDBN model is updated satisfies the following formula: ; in, Represents the current optimized combination index, Represents the last optimized combination index, Represents the preset difference threshold; Specifically, the steps for updating the number of iterations in the model parameters are as follows: First, a preset fluctuation threshold is obtained; then, in consecutive iterations, the average value and standard deviation of the optimized combination index of the iteration are obtained; then, based on the average value and standard deviation of the optimized combination index of the iteration and the preset fluctuation threshold, it is determined whether the CDBN model needs to be iterated further; and, Get the preset fluctuation threshold; In M consecutive iterations, the average value and standard deviation of the optimized combination index of the Mth iteration are obtained; Based on the average and standard deviation of the optimized combination index of the Mth iteration and the preset fluctuation threshold, determine whether to continue iterating the CDBN model; The mean and standard deviation of the optimized combination index of the Mth iteration are obtained by the following formula: ; ; in, Represents the average value of the optimized combination index, represents the standard deviation of the optimized combination index, Represents the optimized combination index of the Mth iteration; The preset fluctuation threshold is obtained by the following formula: ; in, Represents the preset fluctuation threshold, Represents the preset adjustment factor, which is 0.1 here; Based on the average value and standard deviation of the optimized combination index of the Mth iteration and the preset fluctuation threshold, determining whether to continue iterating the CDBN model includes: If the optimization combination index of M consecutive iterations is Fluctuates within a certain range, and ,in, If the difference threshold is exceeded, it is determined that the optimization combination index has no significant downward trend, and the training is stopped, and the number of iterations in the model parameters of the CDBN model is updated with the current number of iterations.

[0036] Based on the recognition result corresponding to each CAD object in the training data set and the classification result corresponding to each CAD object, selectively updating the model parameters of the CDBN model to obtain the updated CDBN model corresponding to the training data set further includes: Based on the optimization combination index of the CDBN model of the training data set, selectively updating the learning rate and the number of iterations in the model parameters of the CDBN model to obtain an updated CDBN model corresponding to the training data set; And re-execute "substituting the training dataset into the constructed CDBN model, extracting features by performing convolution operations on the training dataset layer by layer, and obtaining convolved feature data corresponding to the training dataset" and subsequent steps to obtain two CDBN models corresponding to the two sets of training datasets respectively.

[0037] Specifically, the initial learning rate can be set to 0.001, and the initial number of iterations can be set to 50.

[0038] Specifically, the BIM object type set is For example, for each type , the number of objects of this type extracted in the data preprocessing stage is , in the model training phase, a data set containing objects of this type is used for training, and in the object recognition and classification phase, the number of correctly classified objects of this type is counted as The correct classification rate is .

[0039] Specifically, the preprocessed BIM dataset is substituted into the trained CDBN model to identify and classify multiple BIM data in the OFF format in the preprocessed BIM dataset. The classification results of each OFF format BIM data in the preprocessed BIM dataset include: Obtaining BIM object categories in a preset BIM object database; Based on the preprocessed BIM dataset, determining the trained CDBN model corresponding to the preprocessed BIM dataset; Substitute the preprocessed BIM dataset into the trained CDBN model corresponding to the preprocessed BIM dataset, perform feature extraction on multiple BIM data in OFF format in the preprocessed BIM dataset, and obtain feature data of each BIM data; Based on the feature data of each BIM data, a probability of each category in the preset BIM object category is obtained, thereby determining the category in the preset BIM object category to which each BIM data belongs; The category in the preset BIM object category to which each BIM data belongs is used as the classification result of each OFF format BIM data in the preprocessed BIM dataset.

[0040] Specifically, 76 unique objects selected from the National BIM Library (NBL) and 52 objects extracted from 4 industry BIM models are used as examples. The pre-processed geometric data of these BIM objects are input into the trained CDBN model. Assume that the input BIM object geometric data is represented in 3D space as a The trained CDBN model extracts features by layer-by-layer convolution of the voxel grid data during the recognition process. Suppose the convolution kernel size of the first convolution layer is , the step size is , then the size of the feature map obtained after the first layer of convolution is ; The trained CDBN model outputs an object classification result based on the extracted features, determining which of the predefined BIM object categories the object belongs to. Specifically, in practice, for an unknown BIM object input, the trained CDBN model, after multiple layers of convolution and calculation, outputs a probability of 0.8 that it belongs to the "chair" category, thus determining that the object is a chair.

[0041] In the above embodiment, 76 unique objects were selected from the National BIM Library (NBL) and 52 objects were extracted from 4 industry BIM models as test data. These data cover BIM models of different types and periods, which can comprehensively test the effectiveness of the method of the present invention.

[0042] Step S104: Based on the classification result of each OFF-format BIM data in the preprocessed BIM dataset, semantic data enhancement is selectively performed on each OFF-format BIM data in the preprocessed BIM dataset to obtain a semantically enhanced BIM dataset, so as to improve the reuse efficiency of BIM data in the building operation stage, thereby reducing the data creation cost.

[0043] Specifically, the preset BIM object database stores a plurality of BIM object categories and different types of semantic data matching each BIM object category, and the selectively performing semantic data enhancement on each OFF format BIM data in the preprocessed BIM dataset based on the classification result of each OFF format BIM data in the preprocessed BIM dataset includes: If semantic data is missing from any of the OFF format BIM data in the pre-processed BIM dataset, it is determined that semantic data enhancement needs to be performed on the OFF format BIM data; generating, based on a preset BIM object database and the BIM data in the OFF format, at least one type of semantic data that matches the BIM data in the OFF format; At least one type of semantic data matching the BIM data in the OFF format is added to the BIM data in the OFF format to achieve semantic enhancement of the BIM data.

[0044] Specifically, based on the preset BIM object database and the BIM data in the OFF format, the specific steps of generating at least one type of semantic data matching the BIM data in the OFF format are as follows: Assume that the identified object is , the objects in the BIM object library are , by calculating the Euclidean distance between the two geometric features ,in and Respectively and The value on the jth feature dimension, d is the number of feature dimensions, select the object with the highest similarity The semantic data is added to Specifically, a BIM object is identified as a "table", but the semantic data of this object lacks manufacturer information. By calculating the similarity between the geometric features of other table objects in the NBL, the most similar object is found and its manufacturer information is added to the current object, thus achieving automatic enhancement of semantic data.

[0045] In the above embodiment, BIM objects are accurately identified through geometric data and semantic data is automatically generated, which significantly improves data reuse efficiency and enhances cross-system interoperability.

[0046] Based on the above steps S101 to S104, the acquired BIM data set is preprocessed to obtain a preprocessed BIM data set, and the BIM data is cleaned and formatted uniformly to improve the accuracy and efficiency of data processing by the CDBN model. By substituting the preprocessed BIM data set into the trained CDBN model, the classification result of each BIM data in the preprocessed BIM data set is obtained, and accurate identification and classification of each BIM data is achieved, thereby providing a unified and reliable foundation for cross-system data interaction and supporting efficient collaboration between different systems in the supply chain. Then, based on the classification result of each BIM data, semantic data enhancement of the BIM data is selectively performed, which solves the data reuse problem caused by inconsistent semantic data, improves the reuse efficiency of BIM data in the building operation stage, reduces data creation costs, and provides more accurate and efficient technical means for digital management of the construction industry, which helps to improve the management level of the entire life cycle of construction projects and avoids the problem that BIM technology in the existing technology is difficult to migrate, has poor quality, and is difficult to reuse due to frequent cross-disciplinary data exchange throughout the life cycle of construction projects. The semantic data problem directly restricts BIM. Moreover, semantic data is difficult to use for automatic mining, which makes it difficult to ensure data consistency and accuracy, leading to technical problems such as high processing costs.

[0047] It should be pointed out that although the various steps in the above embodiments are described in a specific order, those skilled in the art will understand that in order to achieve the effects of the present invention, different steps do not have to be performed in such an order. They can be performed simultaneously (in parallel) or in other orders. These changes are within the scope of protection of the present invention.

[0048] Those skilled in the art will appreciate that all or part of the processes in the method for implementing the above-mentioned embodiment of the present invention may also be accomplished by instructing the relevant hardware through a computer program. The computer program may be stored in a computer-readable storage medium. When the computer program is executed by a processor, it may implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which may be in source code form, object code form, executable file, or some intermediate form. The computer-readable storage medium may include: any entity or device, medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory, random access memory, electric carrier signal, telecommunication signal, and software distribution medium capable of carrying the computer program code. It should be noted that the content contained in the computer-readable storage medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable storage media do not include electric carrier signals and telecommunication signals.

[0049] Furthermore, it should be understood that since the configuration of each module is merely for the purpose of illustrating the functional units of the apparatus of the present invention, the physical devices corresponding to these modules may be the processor itself, or a portion of the software in the processor, a portion of the hardware, or a combination of software and hardware. Therefore, the number of modules in the figure is merely illustrative.

[0050] Those skilled in the art will appreciate that the various modules in the device can be adaptively split or merged. Such splitting or merging of specific modules does not cause the technical solution to deviate from the principles of the present invention. Therefore, the technical solutions after splitting or merging will fall within the scope of protection of the present invention.

[0051] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

Claims

1. A method for BIM object recognition and automatic semantic data generation based on deep learning, characterized in that: The method comprises: Acquire a BIM dataset, wherein the BIM dataset includes at least a plurality of BIM data in an IFC format; Preprocessing the BIM dataset to obtain a preprocessed BIM dataset, wherein the preprocessed BIM dataset at least includes a plurality of BIM data in an OFF format and an object category corresponding to each BIM data in the OFF format; Substituting the preprocessed BIM dataset into the trained CDBN model, the multiple BIM data in the OFF format in the preprocessed BIM dataset are identified and classified, and the classification results of each BIM data in the OFF format in the preprocessed BIM dataset are obtained; Furthermore, based on the classification result of each OFF format BIM data in the preprocessed BIM dataset, semantic data enhancement is selectively performed on each OFF format BIM data in the preprocessed BIM dataset to obtain a BIM dataset with enhanced semantic data.

2. The method for BIM object recognition and automatic generation of semantic data based on deep learning according to claim 1, characterized in that: Acquiring the BIM dataset includes: Acquire multiple BIM data from different data sources; The IFC objects in each BIM data are extracted to form multiple BIM data in IFC format to obtain a BIM dataset.

3. The method for BIM object recognition and automatic semantic data generation based on deep learning according to claim 2 is characterized in that: The preprocessing of the BIM dataset to obtain the preprocessed BIM dataset includes: Perform data format conversion on multiple BIM data in IFC format in the BIM dataset to obtain the OFF file corresponding to each BIM data in IFC format; Clean the OFF file corresponding to each BIM data in IFC format to obtain the cleaned OFF file; Based on the IFC categories of multiple BIM data in the IFC format in the BIM dataset, the cleaned OFF files are grouped into categories to obtain the object category corresponding to each OFF format BIM data; Based on the cleaned OFF files and the object categories corresponding to each OFF format BIM data, a preprocessed BIM dataset is constructed.

4. The method for BIM object recognition and automatic semantic data generation based on deep learning according to claim 3 is characterized in that: Before substituting the pre-processed BIM dataset into the trained CDBN model, the method further trains the CDBN model through the following steps: Build the CDBN model; Create two sets of training data sets based on the two preset basic training data sets; The CDBN model is trained based on two sets of training data sets, and two CDBN models are obtained respectively.

5. The method for BIM object recognition and automatic generation of semantic data based on deep learning according to claim 4 is characterized in that: The CDBN model construction includes: Construct multiple CRBM network architectures and stack them to obtain a CDBN model, where each CRBM network architecture includes a visible layer, a hidden layer, and a pooling layer; The CDBN model is expressed by the following formula: ； Where k represents the number of CRBM network architectures, represents the k-th CRBM network architecture, represents the visible layer input data dimension of the k-th CRBM, represents the number of hidden layer neurons of the k-th CRBM, Represents the pooling kernel size of the pooling layer of the k-th CRBM.

6. The method for BIM object recognition and automatic semantic data generation based on deep learning according to claim 5 is characterized in that: The step of creating two sets of training data sets based on the two preset basic training data sets includes: Obtain two preset training data sets, wherein both training data sets include a plurality of CAD object data and an IFC preset category corresponding to each CAD object data; Based on the two preset training data sets, two training data sets are created, wherein both training data sets include a plurality of BIM data in OFF format and an object category corresponding to each BIM data in OFF format.

7. The method for BIM object recognition and automatic semantic data generation based on deep learning according to claim 6, characterized in that: The CDBN model is trained based on two sets of training data sets to obtain two CDBN models respectively, including: Substitute any set of training data sets into the constructed CDBN model, perform convolution operations on the training data sets layer by layer to extract features, and obtain the convolved feature data corresponding to the training data sets; Determining, based on the convolved feature data, a recognition result corresponding to each CAD object in the training data set and a classification result corresponding to each CAD object; Based on the recognition result corresponding to each CAD object in the training data set and the classification result corresponding to each CAD object, selectively updating the model parameters of the CDBN model to obtain an updated CDBN model corresponding to the training data set; Re-execute the steps of "substituting the training dataset into the constructed CDBN model, performing a convolution operation on the training dataset layer by layer to extract features, and obtaining convolved feature data corresponding to the training dataset" and subsequent steps to obtain two CDBN models corresponding to the two sets of training datasets, respectively.

8. The method for BIM object recognition and automatic semantic data generation based on deep learning according to claim 7, characterized in that: The selectively updating the model parameters of the CDBN model based on the recognition result corresponding to each CAD object and the classification result corresponding to each CAD object in the training data set to obtain the updated CDBN model corresponding to the training data set includes: Based on the recognition result corresponding to each CAD object in the training data set and the classification result corresponding to each CAD object, obtaining an optimization combination index of the training data set for the CDBN model; The optimization combination index of the training data set for the CDBN model is obtained by the following formula: ； Among them, L represents the optimization combination index, N represents the number of samples in the training data set, and C represents the number of categories in the IFC preset categories. Represents the actual label of the i-th sample and whether the sample belongs to the c-th category: If it does, then ;otherwise, , represents the probability that the i-th sample belongs to the c-th category predicted by the model; Furthermore, based on the optimized combination index, selectively updating the learning rate and the number of iterations in the model parameters of the CDBN model to obtain an updated CDBN model corresponding to the training data set; Among them, the learning rate in the model parameters of the CDBN model is updated by the following formula: ； in, represents the updated learning rate, represents the learning rate before updating, Represents the preset learning rate attenuation factor, which is taken here ; In addition, the judgment on whether the learning rate in the model parameters of the CDBN model is updated satisfies the following formula: ； in, Represents the current optimized combination index, Represents the last optimized combination index, Represents the preset difference threshold; and, Get the preset fluctuation threshold; In M consecutive iterations, the average value and standard deviation of the optimized combination index of the Mth iteration are obtained; Based on the average and standard deviation of the optimized combination index of the Mth iteration and the preset fluctuation threshold, determine whether to continue iterating the CDBN model; The mean and standard deviation of the optimized combination index of the Mth iteration are obtained by the following formula: ； ； in, Represents the average value of the optimized combination index, represents the standard deviation of the optimized combination index, Represents the optimized combination index of the Mth iteration; The preset fluctuation threshold is obtained by the following formula: ； in, Represents the preset fluctuation threshold, Represents the preset adjustment factor, which is 0.1 here; Based on the average value and standard deviation of the optimized combination index of the Mth iteration and the preset fluctuation threshold, determining whether to continue iterating the CDBN model includes: If the optimization combination index of M consecutive iterations is Fluctuates within a certain range, and ,in, If the difference threshold is exceeded, it is determined that the optimization combination index has no significant downward trend, and the training is stopped, and the number of iterations in the model parameters of the CDBN model is updated with the current number of iterations.

9. The method for BIM object recognition and automatic semantic data generation based on deep learning according to claim 8, characterized in that: The pre-processed BIM dataset is substituted into the trained CDBN model to identify and classify multiple BIM data in the OFF format in the pre-processed BIM dataset. The classification results of each OFF format BIM data in the pre-processed BIM dataset include: Obtaining BIM object categories in a preset BIM object database; Based on the preprocessed BIM dataset, determining the trained CDBN model corresponding to the preprocessed BIM dataset; Substitute the preprocessed BIM dataset into the trained CDBN model corresponding to the preprocessed BIM dataset, perform feature extraction on multiple BIM data in OFF format in the preprocessed BIM dataset, and obtain feature data of each BIM data; Determining, based on the characteristic data of each BIM data, a category among preset BIM object categories to which each BIM data belongs; The category in the preset BIM object category to which each BIM data belongs is used as the classification result of each OFF format BIM data in the preprocessed BIM dataset.

10. The method for BIM object recognition and automatic semantic data generation based on deep learning according to claim 9, characterized in that: The preset BIM object database stores a plurality of BIM object categories and different types of semantic data matching each BIM object category, and the selectively performing semantic data enhancement on each OFF format BIM data in the preprocessed BIM dataset based on the classification result of each OFF format BIM data in the preprocessed BIM dataset includes: If semantic data is missing from any OFF-format BIM data in the pre-processed BIM dataset, it is determined that semantic data enhancement needs to be performed on the OFF-format BIM data; generating, based on a preset BIM object database and the BIM data in the OFF format, at least one type of semantic data that matches the BIM data in the OFF format; At least one type of semantic data matching the BIM data in the OFF format is added to the BIM data in the OFF format to achieve semantic enhancement of the BIM data.

Images