Decoration design construction intelligent digital management method and system based on deep learning

Abstract

The invention relates to the technical field of construction management, and discloses a decoration design construction intelligent digital management method and system based on deep learning, and the method comprises the steps: collecting multi-modal data, carrying out the data cleaning of different types of data in the multi-modal data, sequentially carrying out the cross-modal feature extraction and feature fusion of the cleaned data, and obtaining a multi-modal feature fusion model; obtaining a multi-modal fusion feature matrix; inputting the multi-modal fusion feature matrix into a CNN-Transform hybrid model, and outputting a design optimization scheme and a risk thermodynamic diagram; based on the risk thermodynamic diagram, adaptively adjusting construction procedures and design parameters, and pushing the construction procedures and design parameters to a construction team and a user through an interaction layer; according to the method, subjectivity and hysteresis of manual adjustment are avoided, the problems of process conflicts, reworking and the like are effectively reduced, the project cost is reduced, the project period is shortened, and reliable technical support is provided for digital and intelligent transformation of the decoration industry.

Description

Technical Field

[0001] This invention relates to the field of construction management technology, specifically to a smart digital management method and system for decoration design and construction based on deep learning. Background Technology

[0002] With the digital transformation of the decoration industry, traditional decoration design and construction management models have gradually exposed many pain points. Existing management methods rely heavily on human experience to carry out design optimization, process arrangement and risk prediction, which is inefficient and highly subjective. It is difficult to adapt to the multi-dimensional needs of complex decoration projects and lacks the ability to predict in advance and make dynamic adjustments. At the same time, the adjustment of construction procedures and design parameters is mostly decided manually and cannot be adaptively optimized based on real-time risk data, which easily leads to problems such as process conflicts and poor parameter adaptability. Summary of the Invention

[0003] The purpose of this invention is to solve the above problems by designing a smart digital management method and system for decoration design and construction based on deep learning.

[0004] The first aspect of this invention provides a deep learning-based intelligent digital management method for interior design and construction, the method comprising the following steps: Collect multimodal data, including user requirement text, 3D design model, construction sensor data, and environmental data; Data cleaning is performed on different types of data in multimodal data. Cross-modal feature extraction and feature fusion are then performed on the cleaned data to obtain a multimodal fusion feature matrix. Input the multimodal fusion feature matrix into the CNN-Transformer hybrid model to output the design optimization scheme and risk heatmap; Based on the risk heat map, the construction procedures and design parameters are adaptively adjusted and pushed to the construction team and users through the interaction layer.

[0005] Optionally, in a first implementation of the first aspect of the present invention, the data cleaning for different types of data in multimodal data includes: The BERT pre-trained model is used to semantically encode the user requirement text, and the user requirement text is traversed to remove stop words and redundant expressions. Ambiguous sentences are corrected through context semantic matching, and unstructured text is transformed into structured format to obtain structured requirement text. The PointNet++ algorithm is used to traverse the point cloud data of the 3D design model, filter out abnormal vertices and redundant faces, perform downsampling operation on the remaining point cloud data, and then adjust the topology of the 3D design model through the MeshCNN network. The construction sensor data was analyzed point by point according to the time series. Outliers were identified by combining the isolated forest algorithm and replaced with the mean of the same period. Missing data were filled by time series interpolation and the distribution of environmental data was fitted by Gaussian mixture model.

[0006] Optionally, in a second implementation of the first aspect of the present invention, the step of sequentially performing cross-modal feature extraction and feature fusion on the cleaned data to obtain a multimodal fusion feature matrix includes: The structured requirement text is input into the BERT model for encoding, and then a pooling operation is performed to generate semantic feature vectors. The spatial geometric features of the 3D design model are extracted using the PointNet++ algorithm, and then the spatial feature matrix is ​​generated by mining the node topology relationships using ST-GCN. The LSTM-Attention model was used to extract the temporal feature sequences of the processed construction sensor data and environmental data. Construct an association mapping between semantic feature vectors, spatial feature matrices, and temporal feature sequences, assign attention weights to each feature vector, and generate a multimodal fusion feature matrix.

[0007] Optionally, in a third implementation of the first aspect of the present invention, the step of extracting the spatial geometric features of the 3D design model using the PointNet++ algorithm and then generating a spatial feature matrix by mining the node topology relationships using ST-GCN includes: The FPS algorithm is used to select sampling points from the point cloud data of the 3D design model as the center of the local region, and the local point cloud regions are divided into multiple non-overlapping regions according to a preset radius. For each local point cloud region, the relative coordinates of each vertex within the region are calculated with the region center as the reference. The relative coordinates are then input into the PointNet subnetwork, and the local vertex features are encoded through a multilayer perceptron to obtain the feature vector of each local region. The feature vectors of each local region are fused with the global features of the corresponding region center. Through hierarchical iteration, local geometric features at different scales are integrated to output the spatial geometric features of the 3D design model. Based on the spatial geometric features of nodes, and according to the spatial distance and connection relationship between vertices in the 3D design model, a node adjacency matrix is ​​defined to determine the range of neighboring nodes for each node and to construct a graph data structure. Traverse each node in the graph, aggregate the features of neighboring nodes based on the node adjacency matrix, and perform operations on the current node features and the aggregated neighborhood features using a preset graph convolution kernel to mine the topological relationships between nodes; The output node features are dimensionally integrated and standardized, and a spatial feature matrix is ​​constructed according to the arrangement order of spatial nodes in the 3D design model.

[0008] Optionally, in a fourth implementation of the first aspect of the present invention, the step of inputting the multimodal fusion feature matrix into the CNN-Transformer hybrid model and outputting the design optimization scheme and risk heatmap includes: The multimodal fusion feature matrix is ​​input into the CNN-Transformer hybrid model, and the CNN module is used to extract local spatial features. The local spatial feature map output by the CNN module is flattened into a one-dimensional feature sequence, and the feature sequence is segmented according to a fixed length to obtain the reconstructed feature sequence. The reconstructed feature sequence is input into the Transformer encoder. The multi-head attention mechanism is used to calculate the pairwise similarity of all feature positions in the reconstructed feature sequence, and a corresponding attention weight is assigned to each feature. Then, the feedforward neural network is used to perform linear transformation and non-linear activation processing on the features output by the attention mechanism to output the globally associated features. The local spatial features output by the CNN module and the global correlation features output by the Transformer encoder are combined by concatenation, and a comprehensive feature matrix is ​​generated through adaptive weight allocation. The comprehensive feature matrix is ​​input into two independent fully connected layer branches. One fully connected layer branch performs dimension mapping and sequence parsing on the comprehensive feature matrix to generate feature sequences related to the design optimization scheme that conforms to the decoration design. The other fully connected layer branch constructs a risk representation matrix with spatial dimension based on the feature weight distribution and transforms it into the feature output corresponding to the risk heat map. The output features of the two fully connected layer branches are format-normalized, and the output design optimization scheme and risk heatmap are generated.

[0009] Optionally, in a fifth implementation of the first aspect of the present invention, the extraction of local spatial features using a CNN module includes: The method employs alternating multi-layer convolution, using convolution kernels to slide through the multimodal fusion feature matrix and perform convolution operations on features within local regions. This captures spatial node relationships and local structural features in decoration design and construction. After each convolution layer, a pooling layer is used to downsample the feature map.

[0010] Optionally, in the sixth implementation of the first aspect of the present invention, the step of adaptively adjusting construction procedures and design parameters based on the risk heatmap and pushing them to the construction team and users through the interaction layer includes: Traverse the spatial nodes and risk level labels of the risk heat map, extract the construction links, related procedures and design parameters corresponding to the high risk level, determine the scope of procedures and design parameter types that need to be adjusted, and establish a list of adjustment objects; The population size of gray wolves, the maximum number of iterations, and the search space boundary are set, and the scope of construction procedures and the type of design parameters are used as dimensions of individual populations. Using the risk reduction rate of high-risk areas in the risk heatmap as an indicator, the gray wolf population position is initialized, and α wolves, β wolves, and δ wolves are divided into α wolves, β wolves, and δ wolves according to their fitness values. α wolves guide the population to search in the optimal adjustment direction, while β wolves and δ wolves assist in adjusting the position of individuals in the population. After each iteration, the individual corresponding to the optimal fitness is updated. When the number of iterations reaches the preset maximum value, the parameters corresponding to the current optimal individual are extracted and transformed into specific construction procedure adjustment schemes and design parameter correction values ​​to obtain the optimal adjustment scheme. The optimal adjustment plan is transformed into a visual data format and pushed to the terminals of the construction team and users.

[0011] A second aspect of this invention provides a deep learning-based intelligent digital management system for interior design and construction, the system comprising: The data acquisition module is used to collect multimodal data, which includes user requirement text, 3D design model, construction sensor data and environmental data; The feature extraction module is used to clean different types of data in multimodal data. It sequentially performs cross-modal feature extraction and feature fusion on the cleaned data to obtain a multimodal fusion feature matrix. The solution output module is used to input the multimodal fusion feature matrix into the CNN-Transformer hybrid model and output the design optimization solution and risk heatmap; The push module is used to adaptively adjust construction procedures and design parameters based on the risk heat map, and push them to the construction team and users through the interaction layer.

[0012] A third aspect of the present invention provides a deep learning-based intelligent digital management device for decoration design and construction, the deep learning-based intelligent digital management device for decoration design and construction including a memory and at least one processor, the memory storing instructions; the at least one processor calling the instructions in the memory to cause the deep learning-based intelligent digital management device for decoration design and construction to perform the various steps of the deep learning-based intelligent digital management method for decoration design and construction as described in any of the preceding claims.

[0013] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the steps of the deep learning-based intelligent digital management method for decoration design and construction as described in any of the preceding claims.

[0014] The technical solution provided by this invention involves collecting multimodal data, cleaning different types of data within the multimodal data, sequentially extracting and fusing cross-modal features from the cleaned data to obtain a multimodal fusion feature matrix. This multimodal fusion feature matrix is ​​then input into a CNN-Transformer hybrid model, which outputs a design optimization scheme and a risk heatmap. Based on the risk heatmap, construction procedures and design parameters are adaptively adjusted and pushed to the construction team and users through an interactive layer. This invention fully leverages the correlation between different modal data, enabling early prediction and precise location of construction risks, significantly reducing design deviations and the probability of missed risk assessments. It ensures the logic and parameter adaptability of procedure adjustments, avoids the subjectivity and lag of manual adjustments, effectively reduces process conflicts and rework, lowers project costs and timelines, and provides reliable technical support for the digital and intelligent transformation of the decoration industry. Attached Figure Description

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0016] Figure 1 A flowchart illustrating the intelligent digital management method for decoration design and construction based on deep learning provided in this embodiment of the invention; Figure 2 A schematic diagram of the structure of the intelligent digital management system for decoration design and construction based on deep learning provided in an embodiment of the present invention; Figure 3 A schematic diagram of the structure of a deep learning-based intelligent digital management device for decoration design and construction provided in an embodiment of the present invention. Detailed Implementation

[0017] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0018] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1The flowchart of the intelligent digital management method for decoration design and construction based on deep learning provided in this embodiment of the invention includes the following steps: Step 101: Collect multimodal data, which includes user requirement text, 3D design model, construction sensor data and environmental data; In this embodiment, raw information is collected through structured questionnaires and voice input tools, and the voice data is converted into text format to form a raw requirement text dataset. A laser scanning device is used to perform a comprehensive scan of the renovation site to obtain site cloud data. An initial 3D model is constructed using CAD and SketchUp tools, and a model data file containing vertices, faces, and topological relationships is exported. Construction sensor equipment is deployed, and vibration, displacement, and energy consumption sensors are installed at key locations such as walls, water and electricity pipelines, and construction machinery. A fixed acquisition frequency is set to collect raw sensor data in real time and store it in a database. Temperature, humidity, light, and air quality sensors are arranged in different areas within the renovation site, and data is collected on an hourly basis. The collection time and location information are recorded synchronously to form an environmental data sequence.

[0019] Step 102: Perform data cleaning on different types of data in the multimodal data, and sequentially perform cross-modal feature extraction and feature fusion on the cleaned data to obtain a multimodal fusion feature matrix; In this embodiment, a BERT pre-trained model is used to semantically encode the user requirement text, traversing the user requirement text to remove stop words and redundant expressions, correcting ambiguous statements through contextual semantic matching, and transforming unstructured text into a structured format to obtain structured requirement text; the PointNet++ algorithm is used to traverse the point cloud data of the 3D design model, filtering out abnormal vertices and redundant faces, performing downsampling operations on the remaining point cloud data, and then adjusting the topology of the 3D design model through a MeshCNN network; the construction sensor data is analyzed point by point according to the time series, and outliers are identified and replaced with the mean of the same period by combining the isolated forest algorithm, missing data is filled in by temporal interpolation, and the environmental data distribution is fitted by a Gaussian mixture model.

[0020] In this embodiment, the structured requirement text is input into the BERT model for encoding and then pooling is performed to generate semantic feature vectors. The spatial geometric features of the 3D design model are extracted using the PointNet++ algorithm, and then the node topology relationships are mined using ST-GCN to generate a spatial feature matrix. The temporal feature sequences of the processed construction sensor data and environmental data are extracted using the LSTM-Attention model. The association mapping between the semantic feature vectors, spatial feature matrix, and temporal feature sequences is constructed, and attention weights are assigned to each feature vector to generate a multimodal fusion feature matrix.

[0021] In this embodiment, the FPS algorithm is used to select sampling points from the point cloud data of the 3D design model as local region centers, and multiple non-overlapping local point cloud regions are divided according to a preset radius. For each local point cloud region, the relative coordinates of each vertex within the region are calculated based on the region center. The relative coordinates are input into the PointNet sub-network, and the local vertex features are encoded by a multilayer perceptron to obtain the feature vector of each local region. The feature vectors of each local region are fused with the global features of the corresponding region center. Through hierarchical iteration, local geometric features at different scales are integrated to output the spatial geometric features of the 3D design model. Based on the node features of the spatial geometric features and the spatial distance and connection relationship between vertices of the 3D design model, a node adjacency matrix is ​​defined to determine the range of neighboring nodes of each node and construct a graph data structure. Each node in the graph is traversed, and the features of neighboring nodes are aggregated based on the node adjacency matrix. The current node features and the aggregated neighboring features are operated on by a preset graph convolution kernel to mine the topological relationships between nodes. The output node features are dimensionally integrated and standardized, and a spatial feature matrix is ​​constructed according to the arrangement order of spatial nodes in the 3D design model.

[0022] Step 103: Input the multimodal fusion feature matrix into the CNN-Transformer hybrid model, and output the design optimization scheme and risk heatmap; In this embodiment, the multimodal fusion feature matrix is ​​input into the CNN-Transformer hybrid model, and the CNN module is used to extract local spatial features. The local spatial feature map output by the CNN module is flattened into a one-dimensional feature sequence, and the feature sequence is segmented according to a fixed length to obtain the reconstructed feature sequence. The reconstructed feature sequence is input into the Transformer encoder, and the pairwise similarity of all feature positions in the reconstructed feature sequence is calculated through a multi-head attention mechanism. A corresponding attention weight is assigned to each feature, and then the features output by the attention mechanism are linearly transformed and nonlinearly activated through a feedforward neural network to output global correlation features. The local spatial features output by the CNN module and the global correlation features output by the Transformer encoder are integrated by concatenation, and a comprehensive feature matrix is ​​generated through adaptive weight allocation. The comprehensive feature matrix is ​​input into two independent fully connected layer branches. One fully connected layer branch performs dimensional mapping and sequence parsing on the comprehensive feature matrix to generate a feature sequence related to the design optimization scheme that conforms to the decoration design. The other fully connected layer branch constructs a spatial dimension risk representation matrix based on the feature weight distribution and converts it into the feature output corresponding to the risk heat map. The output features of the two fully connected layer branches are format-standardized to output the design optimization scheme and the risk heat map.

[0023] In this embodiment, a multi-layer convolutional layer is used to run alternately. The multimodal fusion feature matrix is ​​traversed by the convolutional kernel, and the features in the local area are convolved to capture the spatial node association and local structural features in the decoration design and construction. After each convolutional layer, the feature map is downsampled by the pooling layer.

[0024] Step 104: Based on the risk heat map, adaptively adjust the construction procedures and design parameters, and push them to the construction team and users through the interaction layer.

[0025] In this embodiment, the spatial nodes and risk level labels of the risk heatmap are traversed to extract the construction links, related procedures, and design parameters corresponding to high-risk levels. The scope of procedures and design parameter types that need to be adjusted are determined, and an adjustment object list is established. The gray wolf population size, maximum number of iterations, and search space boundary are set, and the scope of construction procedures and design parameter types are used as the dimensions of individual populations. The position of the gray wolf population is initialized using the risk reduction rate of high-risk areas in the risk heatmap as an indicator. The wolves are divided into α wolves, β wolves, and δ wolves according to their fitness values. The α wolves guide the population to search in the optimal adjustment direction, while the β and δ wolves assist in adjusting the position of individual populations. After each iteration, the individual corresponding to the optimal fitness is updated. When the number of iterations reaches the preset maximum value, the parameters corresponding to the current optimal individual are extracted and converted into specific construction procedure adjustment schemes and design parameter correction values ​​to obtain the optimal adjustment scheme. The optimal adjustment scheme is converted into a visual data format and pushed to the terminals of the construction team and users.

[0026] Please see Figure 2 A schematic diagram of the structure of the intelligent digital management system for decoration design and construction based on deep learning provided in this embodiment of the invention. The system includes: The data acquisition module is used to collect multimodal data, which includes user requirement text, 3D design model, construction sensor data and environmental data; The feature extraction module is used to clean different types of data in multimodal data. It sequentially performs cross-modal feature extraction and feature fusion on the cleaned data to obtain a multimodal fusion feature matrix. The solution output module is used to input the multimodal fusion feature matrix into the CNN-Transformer hybrid model and output the design optimization solution and risk heatmap; The push module is used to adaptively adjust construction procedures and design parameters based on the risk heat map, and push them to the construction team and users through the interaction layer.

[0027] Figure 3This is a schematic diagram of the structure of a deep learning-based intelligent digital management device for decoration design and construction provided in an embodiment of the present invention. This deep learning-based intelligent digital management device 300 for decoration design and construction can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 310 (e.g., one or more processors) and a memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) for storing application programs 333 or data 332. The memory 320 and storage media 330 can be temporary or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the diagram), each module may include a series of instruction operations on the deep learning-based intelligent digital management device 300 for decoration design and construction. Furthermore, the processor 310 may be configured to communicate with the storage media 330 and execute a series of instruction operations in the storage media 330 on the deep learning-based intelligent digital management device 300 for decoration design and construction to implement the method provided in the above embodiment.

[0028] The deep learning-based intelligent digital management device 300 for interior design and construction may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3 The structure of the intelligent digital management equipment for decoration design and construction based on deep learning shown does not constitute a limitation on the computer equipment provided by this invention. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0029] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the various steps of the deep learning-based intelligent digital management method for decoration design and construction provided in the above embodiments.

[0030] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described equipment or apparatus / unit can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0031] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0032] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended technical solutions and their equivalents.

Claims

1. A deep learning-based intelligent digital management method for decoration design and construction, characterized in that: The method includes the following steps: Collect multimodal data, including user requirement text, 3D design model, construction sensor data, and environmental data; Data cleaning is performed on different types of data in multimodal data. Cross-modal feature extraction and feature fusion are then performed on the cleaned data to obtain a multimodal fusion feature matrix. Input the multimodal fusion feature matrix into the CNN-Transformer hybrid model to output the design optimization scheme and risk heatmap; Based on the risk heat map, the construction procedures and design parameters are adaptively adjusted and pushed to the construction team and users through the interaction layer.

2. The intelligent digital management method for decoration design and construction based on deep learning as described in claim 1, characterized in that, The data cleaning process for different types of data in multimodal data includes: The BERT pre-trained model is used to semantically encode the user requirement text, and the user requirement text is traversed to remove stop words and redundant expressions. Ambiguous sentences are corrected through context semantic matching, and unstructured text is transformed into structured format to obtain structured requirement text. The PointNet++ algorithm is used to traverse the point cloud data of the 3D design model, filter out abnormal vertices and redundant faces, perform downsampling operation on the remaining point cloud data, and then adjust the topology of the 3D design model through the MeshCNN network. The construction sensor data was analyzed point by point according to the time series. Outliers were identified by combining the isolated forest algorithm and replaced with the mean of the same period. Missing data were filled by time series interpolation and the distribution of environmental data was fitted by Gaussian mixture model.

3. The intelligent digital management method for decoration design and construction based on deep learning as described in claim 2, characterized in that, The process of sequentially extracting and fusing cross-modal features from the cleaned data to obtain a multimodal fusion feature matrix includes: The structured requirement text is input into the BERT model for encoding, and then a pooling operation is performed to generate semantic feature vectors. The spatial geometric features of the 3D design model are extracted using the PointNet++ algorithm, and then the spatial feature matrix is ​​generated by mining the node topology relationships using ST-GCN. The LSTM-Attention model was used to extract the temporal feature sequences of the processed construction sensor data and environmental data. Construct an association mapping between semantic feature vectors, spatial feature matrices, and temporal feature sequences, assign attention weights to each feature vector, and generate a multimodal fusion feature matrix.

4. The intelligent digital management method for decoration design and construction based on deep learning as described in claim 3, characterized in that, The process of extracting spatial geometric features of the 3D design model using the PointNet++ algorithm, and then generating a spatial feature matrix by mining node topological relationships using ST-GCN, includes: The FPS algorithm is used to select sampling points from the point cloud data of the 3D design model as the center of the local region, and the local point cloud regions are divided into multiple non-overlapping regions according to a preset radius. For each local point cloud region, the relative coordinates of each vertex within the region are calculated with the region center as the reference. The relative coordinates are then input into the PointNet subnetwork, and the local vertex features are encoded through a multilayer perceptron to obtain the feature vector of each local region. The feature vectors of each local region are fused with the global features of the corresponding region center. Through hierarchical iteration, local geometric features at different scales are integrated to output the spatial geometric features of the 3D design model. Based on the spatial geometric features of nodes, and according to the spatial distance and connection relationship between vertices in the 3D design model, a node adjacency matrix is ​​defined to determine the range of neighboring nodes for each node and to construct a graph data structure. Traverse each node in the graph, aggregate the features of neighboring nodes based on the node adjacency matrix, and perform operations on the current node features and the aggregated neighborhood features using a preset graph convolution kernel to mine the topological relationships between nodes; The output node features are dimensionally integrated and standardized, and a spatial feature matrix is ​​constructed according to the arrangement order of spatial nodes in the 3D design model.

5. The intelligent digital management method for decoration design and construction based on deep learning as described in claim 1, characterized in that, The process of inputting the multimodal fusion feature matrix into the CNN-Transformer hybrid model and outputting the design optimization scheme and risk heatmap includes: The multimodal fusion feature matrix is ​​input into the CNN-Transformer hybrid model, and the CNN module is used to extract local spatial features. The local spatial feature map output by the CNN module is flattened into a one-dimensional feature sequence, and the feature sequence is segmented according to a fixed length to obtain the reconstructed feature sequence. The reconstructed feature sequence is input into the Transformer encoder. The multi-head attention mechanism is used to calculate the pairwise similarity of all feature positions in the reconstructed feature sequence, and a corresponding attention weight is assigned to each feature. Then, the feedforward neural network is used to perform linear transformation and non-linear activation processing on the features output by the attention mechanism to output the globally associated features. The local spatial features output by the CNN module and the global correlation features output by the Transformer encoder are combined by concatenation, and a comprehensive feature matrix is ​​generated through adaptive weight allocation. The comprehensive feature matrix is ​​input into two independent fully connected layer branches. One fully connected layer branch performs dimension mapping and sequence parsing on the comprehensive feature matrix to generate feature sequences related to the design optimization scheme that conforms to the decoration design. The other fully connected layer branch constructs a risk representation matrix with spatial dimension based on the feature weight distribution and transforms it into the feature output corresponding to the risk heat map. The output features of the two fully connected layer branches are format-normalized, and the output design optimization scheme and risk heatmap are generated.

6. The intelligent digital management method for decoration design and construction based on deep learning as described in claim 5, characterized in that, The extraction of local spatial features using a CNN module includes: The method employs alternating multi-layer convolution, using convolution kernels to slide through the multimodal fusion feature matrix and perform convolution operations on features within local regions. This captures spatial node relationships and local structural features in decoration design and construction. After each convolution layer, a pooling layer is used to downsample the feature map.

7. The intelligent digital management method for decoration design and construction based on deep learning as described in claim 1, characterized in that, The process of adaptively adjusting construction procedures and design parameters based on risk heatmaps and pushing these adjustments to the construction team and users through an interaction layer includes: Traverse the spatial nodes and risk level labels of the risk heat map, extract the construction links, related procedures and design parameters corresponding to the high risk level, determine the scope of procedures and design parameter types that need to be adjusted, and establish a list of adjustment objects; The population size of gray wolves, the maximum number of iterations, and the search space boundary are set, and the scope of construction procedures and the type of design parameters are used as dimensions of individual populations. Using the risk reduction rate of high-risk areas in the risk heatmap as an indicator, the gray wolf population position is initialized, and α wolves, β wolves, and δ wolves are divided into α wolves, β wolves, and δ wolves according to their fitness values. α wolves guide the population to search in the optimal adjustment direction, while β wolves and δ wolves assist in adjusting the position of individuals in the population. After each iteration, the individual corresponding to the optimal fitness is updated. When the number of iterations reaches the preset maximum value, the parameters corresponding to the current optimal individual are extracted and transformed into specific construction procedure adjustment schemes and design parameter correction values ​​to obtain the optimal adjustment scheme. The optimal adjustment plan is transformed into a visual data format and pushed to the terminals of the construction team and users.

8. A deep learning-based intelligent digital management system for interior design and construction, characterized in that: The system includes: The data acquisition module is used to collect multimodal data, which includes user requirement text, 3D design model, construction sensor data and environmental data; The feature extraction module is used to clean different types of data in multimodal data. It sequentially performs cross-modal feature extraction and feature fusion on the cleaned data to obtain a multimodal fusion feature matrix. The solution output module is used to input the multimodal fusion feature matrix into the CNN-Transformer hybrid model and output the design optimization solution and risk heatmap; The push module is used to adaptively adjust construction procedures and design parameters based on the risk heat map, and push them to the construction team and users through the interaction layer.

9. A deep learning-based intelligent digital management device for decoration design and construction, characterized in that: The deep learning-based intelligent digital management device for decoration design and construction includes a memory and at least one processor. The memory stores instructions. The at least one processor invokes the instructions in the memory to cause the deep learning-based intelligent digital management device for decoration design and construction to perform each step of the deep learning-based intelligent digital management method for decoration design and construction as described in any one of claims 1-7.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement each step of the deep learning-based intelligent digital management method for decoration design and construction as described in any one of claims 1-7.

Images