Intelligent building decoration design method based on deep learning

Abstract

The invention discloses an intelligent building decoration design method based on deep learning, and the method comprises the following steps: S1), employing a monocular camera to shoot pictures of a building at all angles, and obtaining an original picture; s2) extracting keywords, and making cue words; s3) building decoration data collection; s4) performing data screening and preprocessing; s5) customizing a loss function, and training the diffusion model to obtain a trained diffusion model; s6) inputting the original image into the trained diffusion model to generate a preliminary image; inputting the preliminary image and the cue word into a pre-trained conditional generative adversarial network to obtain a preliminary design drawing; and S7) inputting the preliminary design drawing into a pre-trained variational auto-encoder to obtain an architectural decoration design reference draw.According to the intelligent architectural decoration design method based on deep learning, through the design of the self-defined loss function and the dynamic weight, in combination with the deep learning model, the image is automatically processed, and the design reference drawing is generated, so that the working efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the field of architectural decoration design, and in particular relates to an intelligent architectural decoration design method based on deep learning. Background Art

[0002] With the continuous emergence of new technologies and new materials, designers need to constantly learn and update their knowledge to adapt to industry development, which distracts designers' attention and affects design efficiency. At the same time, due to work pressure and time constraints, designers' time for innovative thinking is compressed, resulting in a lack of novelty in design works. With the intensification of market competition, designers face the pressure of providing more design solutions in a shorter time, which affects the quality of the design.

[0003] At the same time, the emergence of deep learning technology has improved design efficiency. It can quickly analyze large amounts of data to provide valuable insights and help make more scientific design decisions. Summary of the invention

[0004] Purpose of the invention: In order to overcome the above shortcomings, the purpose of the present invention is to provide an intelligent architectural decoration design method based on deep learning. This method improves the generalization ability of the model through the design of customized loss functions and dynamic weights, avoids over-optimization of specific types of losses, and accelerates the training convergence speed; at the same time, it uses diffusion models and pre-trained conditional generative adversarial networks to provide designers with innovative design solutions; finally, through variational autoencoder technology, the quality of the generated design drawings in terms of clarity and completeness is ensured, thereby improving the efficiency of architectural decoration design.

[0005] Technical solution: In order to achieve the above-mentioned purpose, the present invention provides an intelligent building decoration design method based on deep learning, comprising the following steps: S1): Use a monocular camera to take pictures of the building from various angles to obtain an original image; S2): Extract keywords and make prompt words according to customer needs; S3): Collect relevant building decoration data according to keywords; S4): Screen and preprocess the data to obtain a training set; S5): Customize the loss function and use the training set to train the diffusion model to obtain a trained diffusion model; S6): Input the original image into the trained diffusion model and generate a preliminary image; Input the preliminary image and the prompt words into a pre-trained conditional generative adversarial network to obtain a preliminary design drawing; S7): Input the preliminary design drawing into a pre-trained variational autoencoder to improve the clarity and completeness of the preliminary design drawing, and obtain a reference drawing for architectural decoration design to assist designers in designing.

[0006] Furthermore, the custom loss function in S5) combines the mean square error loss (MSE Loss) and the Dice loss, and sets a dynamic weight at the same time. The specific process includes:

[0007] S501): define a basic loss function;

[0008] S502): Setting the change rate of loss value;

[0009] S503): Adjust weight;

[0010] S504): Calculate custom loss;

[0011] The specific operation of defining the basic loss function in S501) is as follows:

[0012] The mean square error loss calculation formula is as follows:

[0013]

[0014] Where Q is the total number of samples; y i is the true label (target value) of the i-th sample; is the predicted value of the i-th sample;

[0015] The Dice loss calculation formula is as follows:

[0016]

[0017] Among them, y i is the true label (target value) of the i-th sample; is the predicted value of the i-th sample;

[0018] The specific operation of setting the change rate of the loss value in S502) is as follows:

[0019] Calculate the rate of change of loss value between two consecutive training steps; assume that the mean square error loss value at time step t is MSE t , Dice loss value is Dice t , then the rate of change formula is as follows:

[0020] The rate of change of the mean square error loss is:

[0021]

[0022] The rate of change of Dice loss is:

[0023]

[0024] The specific formula for adjusting the weight in S503) is as follows:

[0025]

[0026] λ2=1-λ1

[0027] Among them, λ1 and λ2 are the dynamic weights of mean square error loss and Dice loss respectively, and k is an adjustment factor used to control the sensitivity of weight changes;

[0028] The specific operation of calculating the custom loss in S504) is as follows:

[0029] The custom loss function formula is as follows:

[0030] Combined Loss=λ1·MSE Loss+λ2·Dice Loss

[0031] Among them, λ1 and λ2 are dynamic weights used to balance the mean square error loss and Dice loss. The design of the custom loss function improves the generalization ability of the model, avoids over-optimization of specific types of losses, and speeds up the training convergence speed; at the same time, the dynamic weights can be automatically adjusted according to the performance of the model during the training process, so that the model can better adapt to different data distributions and changes.

[0032] Furthermore, the keyword extraction in S2) specifically uses the TextRank algorithm to extract important sentences or keywords in the text, wherein the text provided is the needs of the customer;

[0033] First, obtain word vectors through the pre-trained word embedding model;

[0034] Next, the relationship between nodes is determined by cosine similarity, and a threshold is set. If the threshold is exceeded, an edge is established; the cosine similarity formula is:

[0035]

[0036] Among them, M and N are two word vectors; θ is the angle between M and N;

[0037] Then, set the initial importance score for each node, usually set to an equal value, and update the importance score of the node itself according to the importance of the current node and the node connected to it; the formula is as follows:

[0038]

[0039] Among them, B is a node connected to node A by an edge; d is the damping factor, which is usually 0.85; Score(B) is the current importance score of node B; OutDegree(B) is the out-degree of node B, which indicates the number of edges connected to node B.

[0040] Finally, wait for the calculation to converge, and get the keywords according to the importance score. The TextRank algorithm does not require pre-learning and training of related documents. It is concise and effective and can be applied to various text data.

[0041] Furthermore, the specific operation of the diffusion model training in S3) is as follows:

[0042] In the forward diffusion process of the diffusion model, the model gradually adds noise to the original data and finally converts it into a standard normal distribution. The process of gradually adding noise can be regarded as a Markov chain, in which each state corresponds to a version of the data with gradually increasing noise; the formula of the Markov chain is as follows:

[0043]

[0044] Among them, x t is the data state at time step t; β t is the noise ratio parameter at time t; ∈ t is the noise sampled from the standard normal distribution N(0,I) at time t; the time step t is an integer from 1 to T, T is the time step, and time T is pure noise;

[0045] The network structure selected in the reverse process is U-Net, the time step and noise ratio parameters are set, a custom loss function (Combined Loss) is used, and the Adam optimization algorithm is used for optimization. The diffusion model can capture the complex structure and statistical characteristics of the data by learning the distribution of the data, so as to realize the subsequent generation of the design diagram.

[0046] Furthermore, the specific operation of inputting the original image into the trained diffusion model and generating the preliminary image in S6) is: performing cropping and normalization processing on the original image as the input of the trained diffusion model; performing reverse diffusion on the trained diffusion model to iteratively generate the first preliminary image. Cropping and normalization processing standardize the image to ensure that the important parts of the image are retained, which is helpful for subsequent iterative generation.

[0047] Furthermore, the specific operation of inputting the preliminary image and the prompt word into the pre-trained conditional generative adversarial network in S6) to obtain the preliminary design diagram is as follows:

[0048] S601): Importing a pre-trained conditional generative adversarial network model and pre-trained weights, and setting the model to evaluation mode;

[0049] S602): Read the first preliminary image, and adjust the preliminary image to the input size required by the pre-trained conditional generative adversarial network model to obtain a second preliminary image;

[0050] S603): Normalizing, scaling the pixel values ​​of the second preliminary image to a range used by the pre-trained conditional generative adversarial network model to obtain a third preliminary image;

[0051] S604): Convert the prompt word into a prompt word vector through a pre-trained word embedding model;

[0052] S605): Input the third preliminary image and the prompt word vector into the pre-trained conditional generative adversarial network model to obtain a preliminary design drawing. The prompt word is input as a condition so that the preliminary image can generate a design drawing that is more in line with a specific style or theme in the pre-trained conditional generative adversarial network model.

[0053] Furthermore, the specific operation of S2) making prompt words is: combining the extracted keywords with the designer's expert prompt words to complete the construction of prompt words. The combination of advantageous keywords and expert prompt words improves the accuracy of prompt words and ensures that the direction of subsequent design drawing generation will not deviate from customer needs.

[0054] Furthermore, the specific operation of S3) collecting relevant building decoration data according to keywords is: collecting relevant pictures from the building decoration public platform through an automated crawler according to the keywords. The automated crawler collects information to make data collection more convenient.

[0055] Furthermore, the specific operation of the S4) data screening and preprocessing to obtain the training set is as follows: the designer reviews the collected pictures, keeps the pictures that meet the customer's needs, and integrates the relevant pictures in the own design drawings to obtain the picture data set; the picture data set is cropped and normalized to obtain the training set. Data screening and preprocessing reduce noise interference during model training and improve the quality of generated images.

[0056] It can be seen from the above technical solution that the present invention has the following beneficial effects:

[0057] 1. The intelligent architectural decoration design method based on deep learning of the present invention generates preliminary design drawings through automated image processing and design, which can reduce the labor of designers and improve work efficiency.

[0058] 2. The intelligent building decoration design method based on deep learning of the present invention designs a custom loss function and sets a dynamic weight for the custom loss function, thereby improving the generalization ability of the model and avoiding over-optimization of a specific type of loss; accelerating convergence, so that the model converges to the optimal solution faster, and improving the model training speed;

[0059] 3. The intelligent building decoration design method based on deep learning of the present invention can generate design drawings that meet personalized needs through keywords and demand production prompts fed back by customers, thereby improving customer satisfaction.

[0060] 4. The present invention provides an intelligent building decoration design method based on deep learning, which generates a preliminary design drawing through a diffusion model and a pre-trained conditional generative adversarial network, and provides an innovative design solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 A flowchart of an intelligent building decoration design method based on deep learning according to the present invention;

[0062] Figure 2 A flowchart of a custom loss function for an intelligent building decoration design method based on deep learning according to the present invention;

[0063] Figure 3 A flowchart for model training of an intelligent building decoration design method based on deep learning according to the present invention;

[0064] Figure 4 This is a detailed flowchart of step S6 of the deep learning-based intelligent building decoration design method described in the present invention. DETAILED DESCRIPTION

[0065] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0066] Example 1

[0067] In this embodiment, Figure 1 The present invention discloses an intelligent building decoration design method based on deep learning, comprising the following steps:

[0068] S1): Use a monocular camera to take pictures of the building from various angles to obtain the original picture;

[0069] S2): Extract keywords and create prompt words according to customer needs;

[0070] S3): Collect relevant architectural decoration data based on keywords;

[0071] S4): Data screening and preprocessing to obtain a training set;

[0072] S5): Customize the loss function and use the training set to train the diffusion model to obtain the trained diffusion model;

[0073] S6): input the original image into the trained diffusion model and generate a preliminary image; input the preliminary image and the prompt word into the pre-trained conditional generative adversarial network to obtain a preliminary design diagram;

[0074] S7): Input the preliminary design drawing into the pre-trained variational autoencoder to improve the clarity and completeness of the preliminary design drawing, and obtain a reference drawing for architectural decoration design to assist designers in designing.

[0075] Example 2

[0076] On the basis of Example 1, in this embodiment, the present invention discloses an intelligent building decoration design method based on deep learning, such as Figure 1 and Figure 2 The custom loss function in S5) combines the mean square error loss (MSE Loss) and the Dice loss, and sets the dynamic weight at the same time. The specific process includes:

[0077] S501): define a basic loss function;

[0078] S502): Setting the change rate of loss value;

[0079] S503): Adjust weight;

[0080] S504): Calculate custom loss;

[0081] The specific operation of defining the basic loss function in S501) is as follows:

[0082] The mean square error loss calculation formula is as follows:

[0083]

[0084] Where Q is the total number of samples; y i is the true label (target value) of the i-th sample; is the predicted value of the i-th sample;

[0085] The Dice loss calculation formula is as follows:

[0086]

[0087] Among them, y i is the true label (target value) of the i-th sample; is the predicted value of the i-th sample;

[0088] The specific operation of setting the change rate of the loss value in S502) is as follows:

[0089] Calculate the rate of change of loss value between two consecutive training steps; assume that the mean square error loss value at time step t is MSE t , Dice loss value is Dice t , then the rate of change formula is as follows:

[0090] The rate of change of the mean square error loss is:

[0091]

[0092] The rate of change of Dice loss is:

[0093]

[0094] The specific formula for adjusting the weight in S503) is as follows:

[0095]

[0096] λ2=1-λ1

[0097] Among them, λ1 and λ2 are the dynamic weights of mean square error loss and Dice loss respectively, and k is an adjustment factor used to control the sensitivity of weight changes;

[0098] In particular, k can be set to 0.1 at the beginning of training to prevent over-rapid fitting; subsequent training is adjusted based on the fitting situation;

[0099] The specific operation of calculating the custom loss in S504) is as follows:

[0100] The custom loss function formula is as follows:

[0101] Combined Loss=λ1·MSE Loss+λ2·Dice Loss

[0102] Among them, λ1 and λ2 are dynamic weights used to balance the mean square error loss and Dice loss.

[0103] In this embodiment, if Figure 1 and Figure 3 , the S2) keyword extraction specifically uses the TextRank algorithm to extract important sentences or keywords in the text, wherein the provided text is the customer's needs;

[0104] First, obtain word vectors through the pre-trained word embedding model;

[0105] Next, the relationship between nodes is determined by cosine similarity, and a threshold is set. If the threshold is exceeded, an edge is established; the cosine similarity formula is:

[0106]

[0107] Among them, M and N are two word vectors; θ is the angle between M and N;

[0108] Then, set the initial importance score for each node, usually set to an equal value, and update the importance score of the node itself according to the importance of the current node and the node connected to it; the formula is as follows:

[0109]

[0110] Among them, B is a node connected to node A by an edge; d is the damping factor, which is usually 0.85; Score(B) is the current importance score of node B; OutDegree(B) is the out-degree of node B, which indicates the number of edges connected to node B.

[0111] Finally, wait for the calculation to converge and get the keywords by sorting according to the importance scores.

[0112] Specifically, the TextRank algorithm can be implemented by importing the jieba library in the Python package.

[0113] In this embodiment, if Figure 1 and Figure 3 The specific operation of the diffusion model training in S3) is as follows:

[0114] In the forward diffusion process of the diffusion model, the model gradually adds noise to the original data and finally converts it into a standard normal distribution. The process of gradually adding noise can be regarded as a Markov chain, in which each state corresponds to a version of the data with gradually increasing noise; the formula of the Markov chain is as follows:

[0115]

[0116] Among them, x t is the data state at time step t; β t is the noise ratio parameter at time t; ∈ t is the noise sampled from the standard normal distribution N(0,I) at time t; the time step t is an integer from 1 to T, T is the time step, and time T is pure noise;

[0117] The network structure selected in the reverse process is U-Net, the time step and noise ratio parameters are set, a custom loss function (Combined Loss) is used, and the Adam optimization algorithm is used for optimization.

[0118] Specifically, the top time step T is set to 2000, and the noise ratio parameter is set to increase linearly with the time step T. The formula is as follows:

[0119]

[0120] Where t is the current time step;

[0121] Specifically, you can build a diffusion model by importing the diffusers package of the Hugging Face platform into the local Python environment.

[0122] In this embodiment, if Figure 1 and Figure 3 The specific operation of S6) inputting the original image into the trained diffusion model and generating the preliminary image is: cropping and normalizing the original image as the input of the trained diffusion model; performing reverse diffusion on the trained diffusion model to iteratively generate the first preliminary image.

[0123] Specifically, the cropping and normalization operations can be implemented by importing the Pillow library of the Python package.

[0124] In this embodiment, if Figure 1 , Figure 3 and Figure 4 The specific operation of S6) inputting the preliminary image and the prompt word into the pre-trained conditional generative adversarial network to obtain the preliminary design diagram is as follows:

[0125] S601): Importing a pre-trained conditional generative adversarial network model and pre-trained weights, and setting the model to evaluation mode;

[0126] S602): Read the first preliminary image, and adjust the preliminary image to the input size required by the pre-trained conditional generative adversarial network model to obtain a second preliminary image;

[0127] S603): Normalizing, scaling the pixel values ​​of the second preliminary image to a range used by the pre-trained conditional generative adversarial network model to obtain a third preliminary image;

[0128] S604): Convert the prompt word into a prompt word vector through a pre-trained word embedding model;

[0129] S605): Input the third preliminary image and the prompt word vector into the pre-trained conditional generative adversarial network model to obtain a preliminary design drawing.

[0130] Specifically, after the third preliminary image is input into the pre-trained conditional generative adversarial network model, it is first converted into a tensor format, concatenated with the prompt word vector, and then input into the subsequent generator network to perform the generation operation of the preliminary design drawing.

[0131] In this embodiment, if Figure 1 The specific operation of making the prompt words in S2) is: combining the extracted keywords with the designer's expert prompt words to complete the construction of the prompt words.

[0132] In this embodiment, if Figure 1The specific operation of S3) collecting relevant building decoration data according to keywords is: collecting relevant pictures from the building decoration public platform through an automated crawler according to the keywords.

[0133] Specifically, automated crawlers can be written and implemented by importing libraries such as requests, BeautifulSoup, and Scrapy from Python libraries.

[0134] In this embodiment, if Figure 1 The specific operations of S4) data screening and preprocessing to obtain the training set are as follows:

[0135] The designer reviews the collected images, keeps the images that meet the customer's needs, and integrates the relevant images in his own design drawings to obtain the image data set; the image data set is cropped and normalized to obtain the training set.

[0136] The above description is only a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements without departing from the principle of the present invention, and these improvements should also be regarded as within the protection scope of the present invention.

Claims

1. A deep learning-based intelligent building decoration design method, characterized in that: The steps include: S1): Use a monocular camera to take pictures of the building from various angles to obtain the original picture; S2): Extract keywords and create prompt words according to customer needs; S3): Collect relevant architectural decoration data based on keywords; S4): Data screening and preprocessing to obtain a training set; S5): Customize the loss function and use the training set to train the diffusion model to obtain the trained diffusion model; S6): input the original image into the trained diffusion model and generate a preliminary image; input the preliminary image and the prompt word into the pre-trained conditional generative adversarial network to obtain a preliminary design diagram; S7): Input the preliminary design drawing into the pre-trained variational autoencoder to improve the clarity and completeness of the preliminary design drawing, and obtain a reference drawing for architectural decoration design to assist designers in designing.

2. The intelligent building decoration design method based on deep learning according to claim 1 is characterized in that: The custom loss function in S5) combines the mean square error loss (MSE Loss) and the Dice loss, and sets the dynamic weight at the same time. The specific process includes: S501): define a basic loss function; S502): Setting the change rate of loss value; S503): Adjust weight; S504): Calculate custom loss; The specific operation of defining the basic loss function in S501) is as follows: The mean square error loss calculation formula is as follows: Where Q is the total number of samples; y i is the true label (target value) of the i-th sample; is the predicted value of the i-th sample; The Dice loss calculation formula is as follows: Among them, y i is the true label (target value) of the i-th sample; is the predicted value of the i-th sample; The specific operation of setting the change rate of the loss value in S502) is as follows: Calculate the rate of change of loss value between two consecutive training steps; assume that the mean square error loss value at time step t is MSE t , Dice loss value is Dice t , then the rate of change formula is as follows: The rate of change of the mean square error loss is: The rate of change of Dice loss is: The specific formula for adjusting the weight in S503) is as follows: λ2=1-λ1 Among them, λ1 and λ2 are the dynamic weights of mean square error loss and Dice loss respectively, and k is an adjustment factor used to control the sensitivity of weight changes; The specific operation of calculating the custom loss in S504) is as follows: The custom loss function formula is as follows: Combined Loss=λ1·MSE Loss+λ2·Dice Loss Among them, λ1 and λ2 are dynamic weights used to balance the mean square error loss and Dice loss.

3. The intelligent building decoration design method based on deep learning according to claim 1 is characterized in that: The S2) keyword extraction specifically uses the TextRank algorithm to extract important sentences or keywords in the text, wherein the text provided is the customer's needs; First, obtain word vectors through the pre-trained word embedding model; Next, the relationship between nodes is determined by cosine similarity, and a threshold is set. If the threshold is exceeded, an edge is established; the cosine similarity formula is: Among them, M and N are two word vectors; θ is the angle between M and N; Then, set the initial importance score for each node, usually set to an equal value, and update the importance score of the node itself according to the importance of the current node and the node connected to it; the formula is as follows: Among them, B is a node connected to node A by an edge; d is the damping factor, which is usually 0.85; Score(B) is the current importance score of node B; OutDegree(B) is the out-degree of node B, which indicates the number of edges connected to node B. Finally, wait for the calculation to converge and get the keywords by sorting according to the importance scores.

4. The intelligent building decoration design method based on deep learning according to claim 2 is characterized in that: The specific operation of the diffusion model training in S3) is as follows: In the forward diffusion process of the diffusion model, the model gradually adds noise to the original data and finally converts it into a standard normal distribution. The process of gradually adding noise can be regarded as a Markov chain, in which each state corresponds to a version of the data with gradually increasing noise; the formula of the Markov chain is as follows: Among them, x t is the data state at time step t; β t is the noise ratio parameter at time t; ∈ t is the noise sampled from the standard normal distribution N(0,I) at time t; the time step t is an integer from 1 to T, T is the time step, and time T is pure noise; The network structure selected in the reverse process is U-Net, the time step and noise ratio parameters are set, a custom loss function (Combined Loss) is used, and the Adam optimization algorithm is used for optimization.

5. The intelligent building decoration design method based on deep learning according to claim 1 is characterized in that: The specific operation of S6) inputting the original image into the trained diffusion model and generating the preliminary image is: cropping and normalizing the original image as the input of the trained diffusion model; performing reverse diffusion on the trained diffusion model to iteratively generate the first preliminary image.

6. The deep learning-based intelligent building decoration design method according to claim 5 is characterized in that: The specific operation of inputting the preliminary image and the prompt word into the pre-trained conditional generative adversarial network to obtain the preliminary design diagram is as follows: S601): Importing a pre-trained conditional generative adversarial network model and pre-trained weights, and setting the model to evaluation mode; S602): Read the first preliminary image, and adjust the preliminary image to the input size required by the pre-trained conditional generative adversarial network model to obtain a second preliminary image; S603): Normalizing, scaling the pixel values ​​of the second preliminary image to a range used by the pre-trained conditional generative adversarial network model to obtain a third preliminary image; S604): Convert the prompt word into a prompt word vector through a pre-trained word embedding model; S605): Input the third preliminary image and the prompt word vector into the pre-trained conditional generative adversarial network model to obtain a preliminary design drawing.

7. The intelligent building decoration design method based on deep learning according to claim 2 is characterized in that: The specific operation of making the prompt words in S2) is: combining the extracted keywords with the designer's expert prompt words to complete the construction of the prompt words.

8. The deep learning-based intelligent building decoration design method according to claim 1 is characterized in that: The specific operation of S3) collecting relevant building decoration data according to keywords is: collecting relevant pictures from the building decoration public platform through an automated crawler according to the keywords.

9. The intelligent building decoration design method based on deep learning according to claim 1 is characterized in that: The specific operations of S4) data screening and preprocessing to obtain the training set are as follows: The designer reviews the collected images, keeps the images that meet the customer's needs, and integrates the relevant images in his own design drawings to obtain the image data set; the image data set is cropped and normalized to obtain the training set.

Images