Building waste material resource artistic reutilization and redesign method

Abstract

The invention discloses a method for artistic reutilization and redesign of building waste material resources, and the method constructs a diffusion generation model for artistic redesign based on a building waste multi-modal data set combining vision and language. The model mainly comprises an image text coding module, a diffusion generation module and a request coding module. According to the method, building waste materials serve as elements to be input, multi-modal data is combined with artistic design styles and design requirements, and a brand-new building waste artistic work generation mode is provided through a diffusion generation model. Simulation experiment results show that different types of building waste materials can be flexibly applied through the generative artificial intelligence technology, and corresponding artistic works are generated in combination with various artistic design styles and design requirements. The innovative method not only provides a new thought and tool for artistic creation, but also provides effective technical support for recycling management and sustainable utilization of building waste materials, and is beneficial to further promotion of cyclic utilization of resources and ecological sustainable development.

Description

Technical Field

[0001] The invention belongs to the field of architectural sustainability and art design, and aims to promote the resource and artistic transformation of construction waste materials through the multimodal data set of construction waste and the combination of generative artificial intelligence technology. Based on the conditional diffusion model, this method provides an innovative solution for the resource-based artistic redesign and reuse of construction waste materials. Background Art

[0002] The artistic redesign and reuse of construction waste materials refers to the use of artistic design and transformation to give artistic value to waste materials generated during the construction process and apply them to creation and production activities. This process not only effectively reduces the pollution of construction waste to the environment, but also creates new economic value and social benefits.

[0003] From the perspective of social benefits, the artistic redesign and reuse of construction waste materials can help improve the level of social aesthetics. The originally worthless waste materials are transformed into works of aesthetic value through artistic design, which can improve the city's artistic taste, cultivate the citizens' artistic sentiment, and promote the improvement of the overall social aesthetic ability. In addition, this process can also enhance the social and cultural connotation. By integrating regional culture and historical culture into artistic creation, the artistic redesign of construction waste materials can enrich the city's cultural heritage and enhance cultural taste. At the same time, this method provides a platform for different social groups to participate in artistic creation, promotes the diversified development of society, and enhances social cohesion.

[0004] From the perspective of economic benefits, the redesign and reuse of construction waste materials also have significant advantages. First, it can create new economic growth points. For example, the use of construction waste materials for redesign can give rise to new industries and business models, create jobs and promote economic growth. Secondly, this process can reduce construction costs and achieve cost savings by reducing waste disposal costs. Finally, the artistic redesign of construction waste materials helps to improve resource utilization efficiency and promote the development of a circular economy. Take the Beijing Bird's Nest as an example. A large amount of reinforced concrete waste was used in its construction. Later, these wastes were reused and made into various artworks and landscape decorations, which not only beautified the environment but also saved resources. The construction of the Shanghai World Expo Park also adopted a similar approach. After the World Expo, these waste materials were transformed into theme parks and public facilities, giving the city new vitality.

[0005] In order to further promote the artistic redesign of construction waste materials, it is particularly important to establish a complete waste material recycling system. Classified collection, transportation and treatment of construction waste will provide a basic guarantee for the redesign and reuse of materials. At the same time, strengthening the research and development of related technologies and exploring new recycling processes and technologies are also the core driving force for the development of this field. In this regard, the public's environmental awareness and aesthetic level also need to be improved, and education and publicity activities will play a key role in this process.

[0006] In recent years, the development of generative AI technology has brought new opportunities for the artistic redesign of construction waste materials. Generative AI can automatically generate new content, such as images, text, music, etc., and has achieved success in many fields. Applying this type of technology to the artistic design of construction waste materials can improve design efficiency and reduce costs in many aspects.

[0007] From the perspective of art design, generative AI can significantly improve design efficiency and shorten the development cycle of art products. For example, designers can quickly generate a large number of design solutions through generative AI, and screen and optimize them to accelerate the creation process. In addition, generative AI can also help designers break through inherent thinking and inspire new creative inspiration. It can generate images of waste materials that are completely different from traditional design ideas, enrich design materials, and even create new art forms, such as algorithmic art, data art, etc., broadening people's understanding and cognition of art.

[0008] In terms of the redesign of construction waste materials, generative AI can quickly identify the reuse value of materials and propose optimized design solutions. For example, through generative models, potential design elements in waste materials can be identified and integrated into new works of art. At the same time, generative AI can simulate the effects of different design solutions and help designers choose the best solution, thereby reducing production costs and material waste. More importantly, this technology can also help popularize waste material art, making this design form more attractive and promoting social acceptance of the artistic redesign of waste materials.

[0009] In summary, generative artificial intelligence, as a technology with great potential, is profoundly affecting the artistic redesign and reuse of construction waste materials. In the future, with the continuous advancement of technology, this field will usher in more innovations and challenges, further promoting the sustainable development of construction waste materials. Summary of the invention

[0010] At present, the treatment methods of construction waste are mainly concentrated in landfill and secondary use. Although secondary use has alleviated the pollution of construction waste to the environment to a certain extent, the economic benefits it brings are relatively limited. The landfill treatment method not only fails to make full use of resources, but also causes serious environmental pollution and produces a large amount of greenhouse gases, further aggravating ecological problems. In addition, the present invention makes it possible to transform waste into art by introducing generative AI technology, creating new economic value, and also providing innovative solutions for the treatment of construction waste. The system shows good adaptability and scalability in practical applications, opening up new ways for the resource utilization of construction waste.

[0011] The present invention proposes a method for the artistic reuse and redesign of construction waste materials based on a conditional generative model. The model encodes the input multimodal construction waste materials, combines specific conditional inputs, and uses a generative diffusion model to achieve the resource-based artistic reuse and redesign of construction waste materials according to the conditional input requirements. The method has strong feasibility and high efficiency.

[0012] The core feature of this method for artistic reuse and redesign of construction waste material resources based on a generative diffusion model is that, relying on a multimodal dataset of construction waste, a text encoder and an image encoder are first used to extract and encode the text and image data of construction waste, respectively. Subsequently, the conditional input encoder encodes the design conditions (such as material type, artistic style, etc.). Finally, these text and image features, as well as conditional features, are input into the conditional diffusion model to achieve the artistic redesign and reuse of construction waste materials, achieving the effect of "turning waste into treasure".

[0013] Based on this, the technical solution adopted by the present invention is: a method for artistic reuse and redesign of construction waste material resources, comprising the following steps:

[0014] The description word features and image features of the construction waste text dataset and image dataset are extracted through the text encoder and image encoder respectively;

[0015] Use conditional encoders to encode conditional information (such as material type, art style, etc.);

[0016] Combining the descriptive features, image features and conditional features of construction waste, the diffusion generation model is used to achieve the artistic resource reuse and redesign of construction waste, and output the works of construction waste materials.

[0017] Furthermore, the image encoder encodes the image input I using a pre-trained image encoder f with fixed c parameters. imgCapture the input image features; the text encoder uses a pre-trained text encoder with fixed parameters f text Encode the input text information.

[0018] Specifically, the image encoder encodes the image input I by: given an image I containing k inputs, the image encoder f img Encode each input image separately and get Where N I Indicates the number of tokens contained in each image. Represents the features of the i-th image input, that is, a low-dimensional vector representation. Represents the input image N I The characteristics of a token, I i represents the input image i, f img (I) represents the total feature of the image. The final feature representation of the image input is the sum of the features of k input images, that is,

[0019] Specifically, the text encoder f text Encoding the input text information includes defining Where N T Indicates the number of words in the text information, f text (T) represents the total characteristics of the input image for the text, Indicates the input text Nth T The features of the words, h T Represents the total features of the input image for the text.

[0020] Furthermore, it also includes using a linear mapping function to transform the image feature dimension d I Mapped to text feature dimension d T .

[0021]

[0022] f represents the mapping function, which can be used to I The corresponding dimension vector is transformed into R d The dimension of the vector is equal to the text feature dimension.

[0023] Furthermore, the conditional encoder encodes the conditional text or image through the encoder according to the input conditional text or artistic style image, respectively encodes the text request information or artistic style image using the text or image encoder, and obtains the request feature q by splicing, and then further processes the request feature q using the mirrored trainable CDM encoder, and finally, links the input of the mirrored CDM encoder to the original CDM model. The image encoder uses CNN and the text encoder uses Transformer.

[0024] Furthermore, the diffusion generation model is derived from the noise z t First, based on the conditional representation I,T at each time step t, the sample sequence with less noise can be obtained in turn through denoising, z t-1 ,…,z0; In order to construct a model f with θ as a parameter θ , at each time step t, the training objective function of the extended model aims to characterize z based on Gaussian noise t Solve the problem of model denoising, which is defined as:

[0025]

[0026] represents the objective function to be optimized, θ represents the parameters that the encoder needs to learn, represents the expectation, ∈ represents the noise added in the forward process, z t The table is Gaussian noise, I and T represent image features and text features respectively.

[0027] The model is trained on a large scale, mainly using text information as the main control source, and training the model f θ Right t Perform denoising.

[0028] Specifically, the model f θ It includes an encoder, an intermediate module and a decoder based on residual connection, namely CDMEncoder, CDM Mid and CDM Decoder.

[0029] Furthermore, a modified cross-attention mechanism is defined based on the noise feature z t As a request, with text features h T As a representation of a keyword K and a value V, it can be defined as:

[0030] z t =z t + CrossAtt(Q = z t ,K=V=[h I ,h T ])

[0031] In the formula, [h I ,h T ] represents the concatenation of image and text features, Q represents request, CrossAtt represents the cross-attention mechanism for interactive learning of image and text features, and h I represents the features of the input image, h T Represents the text features corresponding to the input image.

[0032] Furthermore, a multi-task training strategy is included, which uses a pre-trained image generator to align it with the image input information, introduces the request information feature, and the training objective function is further defined as:

[0033]

[0034] y represents the input image text features of construction waste materials, and q represents the input image text conditional input, such as style control and task description.

[0035] Compared with existing treatment methods, the advantage of this invention is that it can efficiently process construction waste artistically through generative artificial intelligence technology, creating new products with artistic value and commercial potential. At the same time, this method not only reduces the environmental pollution caused by landfill, but also promotes the recycling management and sustainable utilization of construction waste, providing an environmentally friendly and innovative solution for waste treatment. Its specific working principle is as follows Figure 1 The advantages of the present invention are:

[0036] (1) Using generative artificial intelligence technology to solve the recycling, reuse and redesign problems of construction waste can achieve the resource artistic transformation of construction waste more quickly and efficiently. Compared with traditional construction waste treatment methods, this technology is more flexible and more efficient.

[0037] (2) By inputting text and image conditions into the conditional encoder, the artistic style of the generation process can be flexibly controlled, so that the artistic transformation of construction waste is not restricted by a specific style and shows strong flexibility and adaptability.

[0038] (3) The application of generative artificial intelligence technology in the reuse and redesign of construction waste resources has effectively shortened the cycle of traditional art design, significantly improved design efficiency, demonstrated innovation and uniqueness, and brought new possibilities for the artistic treatment of waste.

[0039] like Figure 3As shown in the figure, the conditional diffusion model (CDM) is used as the core and a dual encoder-decoder structure is adopted. At the encoding end, the visual-language pre-trained model architecture (such as CLIP, text image encoder) is used to extract the multimodal features of construction waste materials, including physical properties such as material, shape, and texture. The conditional input module also uses the image text encoder to generate conditional embedding vectors for style conditions and text task descriptions. The denoising process of the diffusion model is guided by the conditional vector to gradually refine random noise into works of art. In the inference stage, the model can generate works with artistic value and environmental significance based on the input graphic information of construction waste materials and the target style description, thereby realizing the creative reuse of construction waste. This method based on deep generative models can not only retain the characteristics of the raw materials, but also give them new aesthetic value, providing an innovative solution for the treatment of construction waste. Therefore, these innovations work together to form an efficient, stable, and reliable construction waste artistic treatment system. The overall design of the system not only solves the technical pain points of traditional treatment methods, but also opens up new ways for the high-value utilization of construction waste. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is the forward process of the diffusion model;

[0041] Figure 2 It is the reverse process of the diffusion model;

[0042] Figure 3 Flowchart for the artistic transformation of construction waste;

[0043] Figure 4 Create a flow chart for the construction waste material - brick cross;

[0044] Figure 5 Create artwork for construction waste materials (bricks). DETAILED DESCRIPTION

[0045] 1. Mathematical principles of diffusion model

[0046] As the most typical model of generative AI, the diffusion model [Reference 1] simulates the random diffusion process and gradually transforms random noise into the target data distribution, thereby generating new data samples. The working principle of the diffusion model can be implemented in two stages: the forward process and the reverse process. The forward process mainly starts from the real data sample and continuously adds noise until the data becomes pure Gaussian noise. This process can be regarded as a data damage process. In contrast, the reverse process starts from pure Gaussian noise, and finally restores the real data sample through gradual denoising. This process can be regarded as a data repair process. At present, the diffusion model has been widely used in the generation of image generation, text generation, audio generation and video generation. It should be pointed out that the diffusion model has the following advantages: generating high-quality samples, relatively simple training process, and can be applied to various data types.

[0047] Forward process: Figure 1 As shown, given a set of real data samples x0~q(x0), the forward process adds Gaussian noise to the sample data T times The resulting q process is shown in the figure, where the size of each step is determined by the Gaussian distribution variance hyperparameter To control, where T represents an adjustable parameter. Since each moment t is only related to the moment t-1, the forward process can be defined as the process from data sample x0 to latent variable x T The Markov chain:

[0048]

[0049] In the formula, ——represents the Gaussian distribution variance hyperparameter, and β1<β2<…<β T ; In this process, as t increases, x t Getting closer and closer to pure noise. 1:T represents the total conditional probability after adding noise to the input x0 at each step, q(x t |x t-1 ) represents the conditional probability after denoising, represents Gaussian distribution, I is a unit vector, and T represents the time step. When T→∞, x T represents Gaussian noise, and The choice of β does not matter. t represents the noise level at the t-th time step. Considering that random sampling in the distribution cannot be propagated using reverse gradient, it is necessary to use the re-parameterization technique to make it differentiable [Reference 2]. At the same time, x t Can be defined as

[0050]

[0051] In the formula, at the same time, ∈ represents noise sampling, β i represents the noise level at the tth time step.

[0052] Assuming that from Gaussian distribution Sampling a z, the reparameterization technique can be defined as:

[0053]

[0054] In the formula, z is a variable with randomness and its mean is μ θ The variance is Gaussian distribution. In particular, the parameter μ θ and It can be obtained through a neural network with θ as a parameter. z represents the sample after resampling, μ θ represents the mean of Gaussian parameters, σ θ represents the standard deviation of Gaussian parameters, ⊙ represents multiplication, and ∈ represents Gaussian sampling. The whole process of "sampling" is still gradient-differentiable, and the randomness is transferred to the parameter ∈.

[0055] Reverse process: Figure 2 As shown in the figure, the inverse process of the diffusion model is also called the denoising inference process, which aims to pass the noisy data distribution The original data distribution is restored in . Similarly, this process can be defined as the following Markov chain:

[0056]

[0057] In the formula, μ θ (x t ,t) represents the mean, σ θ (x t ,t) represents the variance, and both the mean and variance are learned by a neural network with parameter θ. θ (x 0:T ) represents the “integral” of all possible paths of data samples starting from pure noise, p(x T ) represents x T The probability, p θ (x t-1 |x t ) indicates that x t Get x t-1 The conditional probability of , t represents the time step. Based on the literature [3], for any β t represents the noise level at time step t. Parameterized μ θ and σ θ is defined as:

[0058]

[0059] ∈ θ Represents the noise size learned by the network.

[0060] 2 Mathematical principles of diffusion model

[0061] The diffusion model is a type of generative model that can transform Gaussian noise into samples from the learning data distribution in an iterative manner through denoising. The present invention further studies the diffusion model for text-to-image generation for artistic redesign and reuse of construction waste based on multimodal construction waste material characteristics. That is, from the noise z t First, based on the conditional representation I,T at each time step t, we can obtain sample sequences with less noise in turn through denoising, such as z t-1 ,…,z0. In order to construct a model f with θ as a parameter θ , at each time step t, the training objective function of the extended model aims to characterize z based on Gaussian noise t To solve the problem of model denoising, it can be defined as:

[0062]

[0063] represents the objective function, θ represents the learnable parameters of the network, represents the expectation,∈,z t , what do I, and T stand for respectively?

[0064] It should be pointed out that the model is trained on a large scale, mainly using text information as the main control source, and training the model f θ Right t Perform denoising. Figure 3 The model framework for artistic redesign and reuse of construction waste is described, which uses image and text encoders to encode input images and text information respectively. The CDM model is mainly used for image generation, while the additional CDM encoder is mainly used to capture information about text generation conditions, that is, request style information and text information. It is worth noting that Figure 3 Only the encoder for a single multimodal construction waste information, such as bricks, is shown. In fact, the model proposed in the present invention can encode different types of construction waste graphic information at the same time.

[0065] Inspired by the UNet structure [4], the model f θ Including encoder, intermediate module and decoder based on residual connection, these models correspond to Figure 3In addition, these models are also based on the ResNet[5] and Transformer[6] structures. At time step t, the conditional diffusion model aims to generate a Gaussian noise z t , encoding conditional text information, and then generating artworks of corresponding styles by understanding the semantics of text information and combining the characteristics of multimodal building waste input. Figure 3 As shown, the context information includes construction waste image input and text information. The significance of designing this module is to introduce construction waste materials so that the model can learn and master the characteristics of different construction waste materials, and define the condition as y = (I, T), where I represents image features and T represents text features. At the same time, the present invention also introduces style control and the task description corresponding to the artistic design of construction waste and uses it as request information q to define the form of output artwork.

[0066] 2.1 Image text information encoding

[0067] like Figure 3 As shown, for the encoding of the image input I, the present invention uses a pre-trained image encoder f with fixed c parameters. img Capture the input image features. Given an image I containing k inputs (e.g., bricks, wood), the image encoder f img Encode each input image separately and get Where N I Indicates the number of tokens contained in each image, and Represents the features of the i-th image input, that is, a low-dimensional vector representation. I Represents the dimension of a single word encoding. Represents the input image N I The characteristics of the token, N I Represents the Nth Token of the image. The final feature representation of the image input is the sum of the features of the k input images, that is, Similarly, in order to better capture the feature information of the text prompt word, the present invention uses a pre-trained text encoder with fixed parameters f text Encode the input text information. In particular, define Where N T represents the number of words in the text message, and Represents the features of the I-th input text. T In addition, a linear mapping function is further used to convert the image feature dimension d I Mapped to text feature dimension d TIn addition, the present invention uses image and text encoders to encode style and text request information respectively, and obtains request features q by splicing, then further processes the request features q using a mirrored trainable CDM encoder, and finally links the input of the mirrored CDM encoder to the original CDM model.

[0068] 2.2 Modified Cross-Attention Mechanism

[0069] Figure 3 In the lower left corner, the modified criss-cross attention mechanism is given in detail. Different from the standard criss-cross attention mechanism in the CDM model through request Q, keyword K and value V, here, the modified criss-cross attention mechanism is defined as taking the noise feature z t As a request, with text features h T As a representation of a keyword K and a value V, it can be defined as:

[0070] z t =z t + CrossAtt(Q = z t ,K=V=[h I ,h T ]) (10)

[0071] In the formula, [h I ,h T ]——represents the concatenation of image and text features. CrossAtt represents the cross attention mechanism, which is used for interactive learning of image and text features. I Represents the characteristics of the input image.

[0072] 2.3 Multi-task training strategy

[0073] A pre-trained image generator is used to align it with the image input information. Introducing the request information feature, formula (9) can be further defined as:

[0074]

[0075] y represents the input image text features of construction waste materials, and q represents the input image text conditional input, such as style control and task description.

[0076] 2.4 Experimental Setup

[0077] All dataset processing experiments in the present invention were completed in Python 3.6 environment and implemented using the deep learning library Pytorch 1.4 installed on the server. When training the deep learning classifier, 8 Nvidia TITANRTX GPUs were used to train the artistic diffusion model of construction waste. The model was pre-trained based on 310,000 synthetic image-text pairs using the image-text retrieval alignment method. Subsequently, the VL-CSW method proposed in reference [7] fine-tuned the model to enable it to learn knowledge related to construction waste materials. When training the CDM module, only the mirror model was trained and the original model was frozen. To encode image and text features, the ViT-L / 14 encoder was used, which has a similar structure to the Transformer [6]. We used the output of the last layer of ViT-L / 14 as the input of the image-text feature. It should be noted that the present invention uses a generative AI model, combined with artistic creation categories (such as oil painting) and creative styles (such as classicism), to achieve the artistic transformation of construction waste. Since the output is an artwork based on construction waste, the present invention does not use traditional evaluation criteria to evaluate the generated results. In particular, through the diverse cross-combination of various art forms (such as oil painting, printmaking, sculpture, Chinese painting, and mixed media) and various artistic styles (such as classicism, romanticism, realism, and photorealism), Figure 4 As shown in the figure, the characteristics of bricks from construction waste can be extracted for secondary artistic creation, which can promote the sustainable reuse of construction waste. Figure 5 shown.

[0078] References

[0079] [1]Rethmeier N,Augenstein IA Primer on Contrastive Pretraining inLanguage Processing:Methods,Lessons Learned&Perspectives[J].ACM Computing Surveys(CSUR),2021.

[0080] [2]Kingma D,Welling M.Auto-encoding variational bayes[J].arXivpreprint arXiv:1312.6114,2013.

[0081] [3]Ho J,Jain A,Abbeel P.Denoising diffusion probabilistic models[C] / / Proceedings of the Advances in neural information processing systems,2020,33:6840-6851.[4]Ronneberger O,Fischer P,Brox T.U-net:Convolutional networks forbiomedical image segmentation[C] / / Proceedings of the Medical Image Computingand Computer-Assisted Intervention,2015:234-241.

[0082] [5]He K,Zhang X,Ren S,et al.Deep residual learning for imagerecognition[C] / / Proceedings of the IEEE conference on computer vision andpattern recognition.2016:770-778.

[0083] [6]Devlin J,Chang M W,Lee K,et al.Bert:Pre-training of deepbidirectional transformers for language understanding[J].arXiv preprintarXiv:1810.04805,2018.

[0085] [7]Sun Y,Gu ZL,and Yang SB.Probing Vision and Language Models forConstruction Waste Material Recognition[J].Automation in Construction,2024:166:105629.

[0087] [8]Radford A,Kim J,Hallacy C,et al.Learning transferable visualmodels from natural language supervision[C] / / Proceedings of the Internationalconference on machine learning.2021:8748-8763。

Claims

1. A method for artistic reuse and redesign of construction waste material resources, characterized in that: The following steps are involved: The description word features and image features of the construction waste text dataset and image dataset are extracted through the text encoder and image encoder respectively; Encode the conditional information using a conditional encoder; Combining the descriptive features, image features and conditional features of construction waste, a diffusion generation model is used to output a work map of construction waste materials.

2. The method for artistic reuse and redesign of construction waste material resources according to claim 1, characterized in that: The image encoder encodes the image input I using a pre-trained image encoder f with fixed c parameters. img Capture the input image features; the text encoder uses a pre-trained text encoder with fixed parameters f text Encode the input text information.

3. The method for artistic reuse and redesign of construction waste material resources according to claim 2 is characterized by: The image encoder encodes the image input I by: given an image I containing k inputs, the image encoder f img Encode each input image separately and get Where N I Indicates the number of tokens contained in each image. represents the features of the i-th image input, Indicates the input image N I The characteristics of a token, I i represents the input image i, f img (I) represents the total feature of the image. The final feature representation of the image input is the sum of the features of k input images, that is, 4. The method for artistic reuse and redesign of construction waste material resources according to claim 2 is characterized by: The text encoder f text Encoding the input text information includes defining Where N T Indicates the number of words in the text information, f text (T) represents the total characteristics of the input image for the text, Indicates the input text Nth T The features of the words, h T Represents the total features of the input image for the text.

5. A method for artistic reuse and redesign of construction waste material resources according to claim 3 or 4, characterized in that: It also includes using a linear mapping function to transform the image feature dimension d I Mapped to text feature dimension d T .

6. The method for artistic reuse and redesign of construction waste material resources according to claim 1 is characterized by: The conditional encoder encodes the conditional text or image through the encoder according to the input conditional text or artistic style image, respectively encodes the text request information or the artistic style image using the text or image encoder, and obtains the request feature q by splicing, then further processes the request feature q using the mirrored trainable CDM encoder, and finally, links the input of the mirrored CDM encoder to the original CDM model.

7. The method for artistic reuse and redesign of construction waste material resources according to claim 1 is characterized by: The diffusion generative model is derived from the noise z t First, based on the conditional representation I, T at each time step t, the sample sequence with less noise can be obtained in turn through denoising, z t-1 ,…,z0; In order to construct a model f with θ as a parameter θ , at each time step t, the training objective function of the extended model aims to characterize z based on Gaussian noise t Solve the problem of model denoising, which is defined as: represents the objective function to be optimized, θ represents the parameters that the encoder needs to learn, represents the expectation, ∈ represents the target image, z t The table is Gaussian noise, I and T represent image features and text features respectively; The model is trained on a large scale, mainly using text information as the main control source, and training the model f θ Right t Perform denoising.

8. The method for artistic reuse and redesign of construction waste material resources according to claim 7 is characterized by: The model f θ It includes an encoder, an intermediate module and a decoder based on residual connection, namely CDM Encoder, CDMMid and CDM Decoder.

9. A method for artistic reuse and redesign of construction waste material resources according to any one of claims 1 to 8, characterized in that: The modified criss-cross attention mechanism is also defined as taking the noise feature z t As a request, with text features h T As a representation of a keyword K and a value V, it can be defined as: z t =z t +CrossAtt(Q=z t ,K=V=[h I ,h T ]) In the formula, [h I ,h T ] represents the concatenation of image and text features, Q represents request, CrossAtt represents the cross-attention mechanism for interactive learning of image and text features, and h I represents the features of the input image, h T Represents the text features corresponding to the input image.

10. The method for artistic reuse and redesign of construction waste material resources according to claim 9, characterized in that: It also includes a multi-task training strategy, which uses a pre-trained image generator to align it with the image input information, introduces request information features, and the training objective function is further defined as: y represents the input construction waste material image text features, and q represents the input image text conditional input.

Images