Building facade design graph generation method, electronic device and storage medium

Abstract

The application provides a building facade design diagram generation method, an electronic device and a computer readable storage medium. The method comprises: obtaining a to-be-designed substrate image; obtaining ecological description information of a target region; identifying the to-be-designed substrate image and the ecological description information by a preset ecological recognition module to obtain first adaptive description information of the to-be-designed substrate image, wherein the first adaptive description information comprises adaptive environment characteristic information and adaptive plant characteristic information of the to-be-designed substrate image; performing segmentation processing on the to-be-designed substrate image according to the adaptive environment characteristic information by a preset mask module to obtain a mask image of the to-be-designed substrate image; and performing facade design according to the mask image, the adaptive environment characteristic information, the adaptive plant characteristic information and the to-be-designed substrate image by a preset ecological design module to obtain a building facade design diagram conforming to the target region. The application can improve the ecological adaptability of a building facade design scheme.

Description

Technical Field

[0001] This application relates to the field of intelligent building design technology, specifically to a method for generating building facade design drawings, an electronic device, and a computer-readable storage medium. Background Technology

[0002] With the advancement of urban ecological construction, building facades, as an important interface of the urban ecosystem, are receiving increasing attention for their ecological functions. Traditional ecological facade design relies on the ecological knowledge and experience of designers, which suffers from high professional knowledge thresholds and low design efficiency. Related technologies have attempted to apply generative artificial intelligence to building facade design; however, the inventors of this application found in the actual research and development process that these facade designs mostly focus on aesthetic style generation and lack adaptability to the ecological environment. Therefore, there is an urgent need for a building facade design scheme with strong ecological adaptability. Summary of the Invention

[0003] This application provides a method for generating architectural facade design drawings, an electronic device, and a computer-readable storage medium, which can improve the ecological adaptability of architectural facade design schemes.

[0004] In a first aspect, this application provides a method for generating architectural facade design drawings, the method comprising: Obtain the matrix image to be designed; Obtain ecological description information of the target area, wherein the ecological description information includes environmental characteristic information of each ecological sample image and plant characteristic information of each plant sample image; By using a preset ecological identification module, the design substrate image and the ecological description information are identified to obtain the first adaptation description information of the design substrate image. The first adaptation description information includes the adaptation environment characteristic information and the adaptation plant characteristic information of the design substrate image. The substrate image to be designed is segmented according to the adaptive environment characteristic information by a preset masking module to obtain a masking image of the substrate image to be designed. The masking image is used to indicate the target segmentation area suitable for plant growth. By using a preset ecological design module, facade design is performed based on the masking image, the adaptive environment characteristic information, the adaptive plant characteristic information, and the substrate image to be designed, resulting in a building facade design drawing that conforms to the target area.

[0005] Secondly, this application also provides an electronic device, which includes a processor and a memory, wherein the memory stores a computer program, and the processor executes any of the building facade design drawing generation methods provided in this application when it calls the computer program in the memory.

[0006] Thirdly, this application also provides a computer-readable storage medium having a computer program stored thereon, the computer program being loaded by a processor to execute the method for generating architectural facade design drawings.

[0007] In this application, firstly, by acquiring ecological description information of the target area and using a preset ecological identification module, the first adaptive description information of the substrate image to be designed is determined based on the ecological description information. This ensures that the first adaptive description information (including adaptive environmental characteristic information and adaptive plant characteristic information) conforms to the environmental and plant characteristics of the target area. Consequently, using this adaptive environmental characteristic information and plant characteristic information for facade design allows for automatic analysis of the local ecological conditions of the target area, improving the ecological adaptability of the building facade design. Secondly, by segmenting the substrate image to be designed based on the adaptive environmental characteristic information to obtain a mask image, and then performing facade design based on the mask image, the suitable target segmentation area for plant growth can be clearly identified, providing regional constraints for ecological design. This ensures that when using the mask image for facade design, the plants in the building facade design are distributed in substrate areas suitable for their growth, improving the ecological adaptability of the building facade design. Therefore, this application can improve the matching degree between the building facade design and the actual ecological needs of the target area, enhancing the ecological adaptability of the building facade design scheme. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a schematic block diagram of the structure of an electronic device provided in an embodiment of this application; Figure 2 This is a schematic flowchart of a method for generating architectural facade design drawings provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structural framework of a preset segmentation model provided in the embodiments of this application; Figure 4 This is a schematic diagram of the structural framework of the ecological design module to be trained in an embodiment of this application; Figure 5 This is a schematic diagram of the structural framework of a preset image generation model provided in the embodiments of this application; Figure 6 This is a schematic diagram of the structural framework of the building facade design model in the embodiments of this application. Detailed Implementation

[0010] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0011] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0012] In the description of the embodiments of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0013] To enable any person skilled in the art to implement and use this application, the following description is provided. In this description, details are set forth for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be implemented without using these specific details. In other instances, well-known processes will not be described in detail to avoid obscuring the description of the embodiments of this application with unnecessary detail. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in the embodiments of this application.

[0014] This application provides a method for generating architectural facade design drawings, an electronic device, and a computer-readable storage medium. The electronic device may be a mobile robot, a mobile phone, a computer, etc.

[0015] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0016] Figure 1 This is a schematic block diagram of the structure of an electronic device provided in an embodiment of this application.

[0017] like Figure 1As shown, the electronic device 100 includes a processor 101 and a memory 102, which are connected by a bus 103, such as a PCIe (Peripheral Component Interconnect Express) bus.

[0018] Specifically, processor 101 provides computing and control capabilities to support the operation of the entire electronic device 100. Processor 101 can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, the general-purpose processor can be a microprocessor or any conventional processor.

[0019] Specifically, the memory 102 can be a Flash chip, a read-only memory (ROM) disk, an optical disk, a USB flash drive, or a portable hard drive, etc.

[0020] Those skilled in the art will understand that Figure 1 The structures shown are merely block diagrams of some structures related to the embodiments of this application and do not constitute a limitation on the electronic devices to which the embodiments of this application are applied. Specific electronic devices may include more or fewer components than those shown in the figures, or combine certain components, or have different component arrangements.

[0021] The processor 101 is configured to run a computer program stored in the memory 102, and when executing the computer program, implement any of the methods for generating architectural facade design drawings provided in this application embodiment. For example, the processor 101 is configured to run a computer program stored in the memory 102, and when executing the computer program, it can perform the following steps: The process involves: acquiring an image of the substrate to be designed; acquiring ecological description information of the target area, wherein the ecological description information includes environmental characteristic information of each ecological sample image and plant characteristic information of each plant sample image; using a preset ecological recognition module, identifying the substrate image to be designed and the ecological description information to obtain first adaptation description information of the substrate image to be designed, wherein the first adaptation description information includes adaptation environmental characteristic information and adaptation plant characteristic information of the substrate image to be designed; using a preset masking module, segmenting the substrate image to be designed based on the adaptation environmental characteristic information to obtain a masked image of the substrate image to be designed, wherein the masked image is used to indicate target segmented areas suitable for plant growth; and using a preset ecological design module, performing facade design based on the masked image, the adaptation environmental characteristic information, the adaptation plant characteristic information, and the substrate image to be designed to obtain a building facade design drawing that conforms to the target area.

[0022] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the electronic device described above can be referred to the corresponding process in the following embodiment of the method for generating architectural facade design drawings, and will not be repeated here.

[0023] The following will be based on Figure 1 Taking the electronic device shown as the execution subject of the method for generating the building facade design drawing as an example, the method for generating the building facade design drawing provided in this application embodiment will be described in detail. For the sake of simplification and ease of description, the execution subject will be omitted in the subsequent method embodiments.

[0024] Please see Figure 2 , Figure 2 This is a flowchart illustrating a method for generating a building facade design drawing according to an embodiment of this application. The method for generating the building facade design drawing includes steps 201-205, wherein: 201. Obtain the matrix image to be designed.

[0025] The substrate image to be designed refers to the substrate image of the building facade to be designed. This image includes the substrate of the building facade but does not include vegetation. In some embodiments, the substrate image to be designed can be an image taken by a user of the building facade. For example, if a user wants to design a facade that conforms to the local ecology for an already constructed building, they can take a picture of the building facade and use the resulting image as the substrate image to be designed. In some embodiments, the substrate image to be designed can also be a sample image of a substrate. For example, a user can select a sample image of a substrate from a building materials website as the substrate image to be designed.

[0026] 202. Obtain ecological description information for the target area.

[0027] The target area refers to the region where the building to be designed is located. The ecological description information must be consistent with the local ecological characteristics of the region to ensure that the subsequently generated building facade design drawings meet the ecological adaptation requirements of the region.

[0028] The ecological description information includes environmental characteristic information for each ecological sample image and plant characteristic information for each plant sample image. Ecological description information refers to the collection containing environmental characteristic information for each ecological sample image and plant characteristic information for each plant sample image; it is the core foundational data for subsequent identification and adaptation of environmental and plant characteristic information.

[0029] Among them, environmental characteristic information is used to describe whether the substrate is suitable for plant growth. For example, environmental characteristic information can be "concrete or stone paving / slabs are not suitable for plant growth. The gaps between the slabs contain soil, dust and organic matter, and have good air permeability, which is suitable for plant growth."

[0030] Among them, plant characteristic information is used to describe the characteristics of plants. For example, plant characteristic information can be "XX plant is a perennial herbaceous plant belonging to the Oxalidaceae family. It is suitable for growing in a warm, humid environment with direct sunlight and has a strong reproductive capacity."

[0031] Among them, ecological sample images refer to matrix images containing plants, which are an important component of ecological description information. Their corresponding environmental characteristic information is used to characterize the ecological adaptation features of this type of matrix.

[0032] For example, ecological data collection is first carried out in the target area where the building to be designed is located. The core data collection includes two types: first, substrate images of plants (i.e., ecological sample images) and their corresponding environmental characteristics; second, real-world images of different plants (i.e., plant sample images) and their corresponding plant characteristics. The collected ecological sample images and plant sample images are standardized, and the environmental characteristics and plant characteristics information are standardized and organized to ensure that the information is presented in a consistent and logical manner. All the organized ecological sample images are associated with their corresponding environmental characteristics, and all the plant sample images are associated with their corresponding plant characteristics, and the information is integrated to form ecological description information. This provides complete and accurate ecological foundation data for subsequent steps, ensuring that the subsequent ecological description information identification and facade design are in line with the local ecological characteristics of the target area.

[0033] For example, if the target area for the building to be designed is Shenzhen, ecological data collection work will be carried out in Shenzhen: First, real-life images of native plants growing on different substrates (concrete, brick, rammed earth, etc.) within the Shenzhen area will be collected, totaling 500 images, as ecological sample images. At the same time, the environmental characteristics information of each ecological sample image will be recorded, such as "the concrete substrate is dense and not suitable for plant growth when there is no soil attached to the surface. The concrete gaps contain soil, dust and organic matter, and have good air permeability, which is suitable for the growth of ground cover plants and mosses" and "the brick substrate is not suitable for plant growth, but the gaps between bricks are suitable for the growth of mosses", etc., to ensure that each ecological sample image corresponds one-to-one with the corresponding environmental characteristics information. Secondly, real-world images of native plants in the Shenzhen area were collected, totaling 9,000 images (including different types such as ground cover, mosses, broadleaf herbs, and ferns) as plant sample images. At the same time, plant characteristic information for each plant sample image was recorded, such as "Wedelia candel is a perennial herb belonging to the Asteraceae family. It is suitable for growing in warm, humid, and sunny environments. It has a strong reproductive capacity, with a plant height of 5-10cm and a creeping growth pattern." "Mosses are suitable for growing in cool, humid, and well-ventilated environments. They grow in an adhesive manner and are suitable for areas such as brick joints and concrete crevices." This ensures that each plant sample image corresponds one-to-one with the corresponding plant characteristic information. The collected 500 ecological sample images and 9,000 plant sample images were standardized and uniformly adjusted to 512×512 pixels and PNG format. The environmental characteristic information and plant characteristic information were standardized and organized, redundant descriptions were removed and professional terminology was unified. All ecological sample images were associated with their corresponding environmental characteristic information and all plant sample images were associated with their corresponding plant characteristic information and stored together to form ecological description information adapted to the Shenzhen area, which will be used for the ecological description information identification of the matrix images to be designed in the future.

[0034] 203. By using a preset ecological identification module, the first adaptation description information of the matrix image to be designed is obtained by identifying the matrix image to be designed and the ecological description information.

[0035] The first adaptation description information includes the adaptation environment characteristics information and the adaptation plant characteristics information of the substrate image to be designed.

[0036] In some embodiments, the adaptation environment characteristic information is identified in the following way: The matrix image to be designed is embedded to obtain a first image embedding vector; based on the first image embedding vector and the image embedding vectors of each of the ecological sample images, a target ecological sample image that matches the matrix of the matrix image to be designed is determined; based on the ecological description information, the environmental characteristic information of the target ecological sample image is obtained as the adaptation environmental characteristic information.

[0037] Among them, the first image embedding vector refers to the high-dimensional vector obtained after embedding the matrix image to be designed. It can characterize the matrix visual features (such as matrix material, texture, structure, etc.) of the matrix image to be designed and is the core basis for matching target ecological sample images.

[0038] Among them, the image embedding vector of the ecological sample image refers to the high-dimensional vector obtained after embedding each ecological sample image, which can characterize the matrix visual features of the corresponding ecological sample image and is used to compare similarity with the first image embedding vector.

[0039] Among them, the target ecological sample image refers to the ecological sample image selected from all ecological sample images that is most similar to the matrix visual features of the matrix image to be designed, and its environmental characteristic information is the adaptive environmental characteristic information of the matrix image to be designed.

[0040] For example, on the one hand, a pre-defined image embedding model (such as Vision Transformer, or ViT, based on the Transformer architecture) is used to encode the matrix image to be designed, obtaining the first image embedding vector. This achieves the digital representation of the matrix image to be designed, providing a computable feature basis for subsequent matching with ecological sample images, ensuring the accuracy and efficiency of the matching process. On the other hand, a pre-defined image embedding model can be used to pre-encode all ecological sample images in the ecological description information, obtaining the image embedding vector corresponding to each ecological sample image, and constructing an ecological sample embedding vector library. Specifically, ViT first divides the image to be encoded into fixed-size image patches, transforms the two-dimensional image patches into a one-dimensional token sequence, adds image spatial location encoding to this sequence to give it location features and adapt it to the sequence processing characteristics of Transformer, and then uses a Transformer encoder to perform feature encoding and fusion of the token sequence with fused location information through a self-attention mechanism. The final output high-dimensional feature vector is the image embedding vector. Then, the similarity (e.g., cosine similarity) between the first image embedding vector and the image embedding vector of each ecological sample image in the vector library is calculated. The ecological sample image with the highest similarity is selected and identified as the target ecological sample image, ensuring that the matrix of the target ecological sample image has the highest visual consistency with the matrix of the matrix image to be designed. The ecological description information stores the association between each ecological sample image and the corresponding environmental characteristic information. After determining the target ecological sample image, the environmental characteristic information corresponding to the target ecological sample image is retrieved from the ecological description information based on this association. Since the matrix of the target ecological sample image is highly matched with the matrix of the matrix image to be designed, its corresponding environmental characteristic information is also adapted to the matrix image to be designed. Therefore, this environmental characteristic information is directly used as the adapted environmental characteristic information of the matrix image to be designed, providing a core basis for subsequent segmentation processing. In this way, by directly reusing the environmental characteristic information corresponding to the target ecological sample image as the adapted environmental characteristic information, there is no need to re-analyze the environmental characteristics of the matrix image to be designed, simplifying the recognition process and ensuring that the adapted environmental characteristic information fits the ecological characteristics of the target area, providing an accurate ecological basis for subsequent segmentation processing and facade design.

[0041] In some embodiments, the adaptive plant characteristic information is identified in the following manner: Obtain the second image embedding vector of the target ecological sample image; based on the second image embedding vector and the image embedding vectors of each plant sample image, determine the target plant sample image that matches the plant in the target ecological sample image; obtain the plant characteristic information of the target plant sample image based on the ecological description information, as the adapted plant characteristic information.

[0042] For example, on the one hand, the target ecological sample image is preprocessed (standardization, size adjustment, etc.). Using a feature extraction layer of a pre-defined image embedding model (such as ViT), the core visual features of the plants in the image (plant species, growth morphology, leaf texture, etc.) are extracted. Then, through the embedding layer of the pre-defined image embedding model, the extracted plant visual features are transformed into a high-dimensional vector with the same dimension as the plant sample image embedding vector, i.e., a second image embedding vector. This achieves a digital representation of the plant features in the target ecological sample image, providing a calculable feature basis for subsequent matching with plant sample images, ensuring the accuracy and consistency of the matching process. On the other hand, using an embedding model and processing standard similar to the pre-defined image embedding model, all plant sample images in the ecological description information are pre-embedded to obtain the image embedding vector corresponding to each plant sample image, constructing a plant sample embedding vector library. Then, the similarity (such as cosine similarity) between the second image embedding vector and the image embedding vector of each plant sample image in the vector library is calculated. The plant sample image with the highest similarity is selected and determined as the target plant sample image, ensuring that the plant features of the target plant sample image have the highest visual consistency with the plant features of the target ecological sample image, thereby adapting to the growth requirements of the substrate to be designed. The ecological description information stores the association between each plant sample image and its corresponding plant characteristic information. After determining the target plant sample image, the plant characteristic information corresponding to the target plant sample image is retrieved from the ecological description information based on this association. Since the plant characteristics of the target plant sample image are highly matched with the plant characteristics of the target ecological sample image, and the matrix of the target ecological sample image is highly matched with the matrix of the matrix image to be designed, the plant characteristic information is adapted to the matrix image to be designed. It is directly used as the adapted plant characteristic information of the matrix image to be designed, providing the core basis for the generation of subsequent facade design prompt information.

[0043] 204. Using a preset masking module, the matrix image to be designed is segmented according to the adaptive environment characteristic information to obtain a masked image of the matrix image to be designed.

[0044] The masked image is used to indicate the target segmentation region suitable for plant growth.

[0045] In some embodiments, step 204 may specifically include the following steps 2041 to 2043: 2041. The matrix image to be designed is segmented through the segmentation layer of the preset masking module to obtain a semantic segmentation map of the matrix image to be designed.

[0046] The semantic segmentation map includes segmentation regions of multiple different matrix types.

[0047] The preset masking module refers to a functional module that, after training, can segment the matrix image to be designed based on the characteristics of the adaptation environment and output a masking image. It includes three functional layers: segmentation layer, adaptation layer, and masking layer.

[0048] Please refer to Figure 3 , Figure 3 This is a schematic diagram of the structural framework of a preset segmentation model provided in this application embodiment. The preset segmentation model includes a segmentation layer, an adaptation layer, and a masking layer. The segmentation layer is used to segment the image of the substrate to be designed, outputting a semantic segmentation map containing segmented regions of multiple different substrate types. The adaptation layer is used to perform ecological adaptation analysis on each segmented region based on adaptation environment characteristic information and the semantic segmentation map, outputting growth classification results. The masking layer is used to mask the image of the substrate to be designed based on the growth classification results, retaining areas suitable for plant growth and masking unsuitable areas to obtain a masked image.

[0049] Among them, the semantic segmentation map refers to the image output by the segmentation layer, which can clearly divide the segmentation regions of different matrix types in the matrix image to be designed and clarify the matrix attributes of each region.

[0050] In this context, a segmented region refers to a region with a single matrix type that is divided in the semantic segmentation graph. Each segmented region corresponds to a specific building facade matrix.

[0051] Among them, the growth classification result is the classification result output by the adaptation layer, which is used to indicate whether the corresponding segmented region is suitable for plant growth, and is the core basis for masking processing.

[0052] For example, the matrix image to be designed is input into the segmentation layer. The segmentation layer performs pixel-level segmentation on the matrix image using a preset semantic segmentation algorithm, identifying different types of matrix in the image (such as concrete, brick, rammed earth, etc.), and dividing pixels of the same matrix type into an independent region. Finally, a semantic segmentation map is output, which clearly presents the segmented regions of multiple different matrix types, providing a clear regional division basis for subsequent ecological adaptation analysis. For example, the matrix image to be designed is a brick facade image of a building in Shenzhen (512×512 pixels, PNG format). This image contains two main matrix types: the brick matrix body and the brick joint gaps. A segmentation layer with a preset masking module is loaded. This segmentation layer uses the U-Net semantic segmentation algorithm. After the matrix image to be designed is input into the segmentation layer, the segmentation layer performs pixel-level recognition on the image, dividing the pixels corresponding to the brick matrix body into one segmented region and the pixels corresponding to the brick joint gaps into another segmented region, labeling the matrix types of the two regions ("brick matrix body" and "brick joint gaps," respectively). Finally, a semantic segmentation map containing these two segmented regions is output, clearly distinguishing the distribution range of different matrix types.

[0053] 2042. Through the adaptation layer of the preset masking module, ecological adaptation analysis is performed based on the adaptation environment characteristic information and the semantic segmentation map to obtain the growth classification results of each segmented region.

[0054] For example, the adaptation environment characteristic information and the semantic segmentation map obtained in step 2041 are input into the adaptation layer of the preset masking module. The adaptation layer first extracts the matrix type information of each segmented region in the semantic segmentation map, and then combines the description of "whether different matrix types are suitable for plant growth" in the adaptation environment characteristic information to analyze each segmented region one by one, determine whether the region is suitable for plant growth, and finally output the growth classification result containing all segmented regions, clearly marking each region as "suitable for growth" or "unsuitable for growth". For example, the adaptation environment characteristic information is "the brick main substrate is not suitable for plant growth, while the substrate in the brick joints contains soil, dust, and organic matter, has good air permeability, and is suitable for plant growth." This adaptation environment characteristic information and the semantic segmentation map obtained in step 2041 (containing two segmentation regions: "brick main substrate" and "brick joints") are input into the adaptation layer. The adaptation layer extracts the substrate type of the two segmentation regions and performs matching analysis with the adaptation environment characteristic information respectively: for the "brick main substrate" segmentation region, it matches "the brick main substrate is not suitable for plant growth" and is judged as "unsuitable for growth"; for the "brick joints" segmentation region, it matches "brick joints are suitable for plant growth" and is judged as "suitable for growth". Finally, the growth classification results are output: brick main substrate segmentation region (unsuitable for growth), brick joints segmentation region (suitable for growth).

[0055] 2043. Using the masking layer of the preset masking module, and based on the growth classification results, the semantic segmentation map is masked to obtain the masked image.

[0056] For example, the masking layer of the preset masking module is loaded, and the semantic segmentation map and the growth classification result obtained in step 2042 are input. The masking layer retains the segmentation regions suitable for plant growth (without masking) and masks the segmentation regions unsuitable for plant growth according to the preset masking rules. The masking image is generated through layer processing. The masking image can intuitively indicate the target segmentation regions suitable for plant growth in the matrix image to be designed, providing a clear basis for the subsequent constraint submodule to determine the plant growth constraint information. For example, the semantic segmentation map (including the segmentation region of the brick body and the segmentation region of the brick joint gap) and the growth classification results (the brick body is unsuitable and the brick joint gap is suitable) are input into the masking layer. The masking layer adopts the preset rule of "white preservation and black masking". The original image pixels (displayed in white) are preserved for the segmentation region of the brick joint gap (suitable for growth), and the segmentation region of the brick body (unsuitable for growth) is masked with black to block the matrix features of the region. After processing, the masked image is output. In this image, only the white area (brick joint gap) is the target segmentation region suitable for plant growth, and the black area (brick body) is the area where plant growth is prohibited, which can clearly indicate the range of subsequent plant growth.

[0057] 205. Using the preset ecological design module, facade design is performed based on the first adaptation description information and the matrix image to be designed, to obtain a building facade design drawing that conforms to the target area.

[0058] In some embodiments, the preset ecological design module includes a prompt submodule, a constraint submodule, and a generation submodule. Step 205 may specifically include the following steps 2051-2053: 2051. Through the prompting submodule, the structured information of the adaptive environment characteristics and the adaptive plant characteristics is obtained by performing structured processing.

[0059] There are several ways to implement step 2051, including, for example: (1) In some embodiments, the facade design prompt information is a structured description of the substrate type and plant type. In this case, step 2051 may specifically include the following steps A1~A3: A1. Extract matrix-related information from the adaptive environment characteristic information, and extract plant category information from the adaptive plant characteristic information.

[0060] A2. According to the preset prompt word structuring rules, the matrix-related information and the plant category information are combined and encoded to form multi-dimensional basic prompt information.

[0061] The plant category information includes, but is not limited to, at least one of herbs, ferns, mosses, succulents, ground cover, and broadleaf herbs.

[0062] A3. Perform semantic optimization and hierarchical division on the basic prompt information to generate facade design prompt information that includes matrix type and plant type.

[0063] (2) In some embodiments, the facade design prompt information is a structured description of the substrate type and plant type. In this case, step 2051 may specifically include the following steps B1~B2: B1. Extract matrix-related information from the adaptive environment characteristic information, and extract plant category information and plant growth morphology information from the adapted plant characteristic information.

[0064] For example, the matrix-related information "brick joint matrix" can be extracted from the environmental characteristic information "brick main matrix is ​​not suitable for plant growth, the matrix in the brick joint gap contains soil, dust and organic matter, has good air permeability and is suitable for plant growth", and the plant category information "broadleaf herb" and the plant growth morphology information "growing along the matrix surface" can be extracted from the plant characteristic information "native herbaceous plants that grow close to the ground, broadleaf native herbaceous plants".

[0065] B2. According to the preset prompt word structure rules, the matrix-related information, plant category information, and plant growth morphology information are combined and encoded to form facade design prompt information that includes suitable matrix type, suitable plant type, and suitable growth morphology.

[0066] The plant species information includes, but is not limited to, at least one of the following: herbs (such as broadleaf herbs), ferns, mosses, succulents, and ground cover.

[0067] Among them, the preset prompt word structured rules refer to the fixed rules that are pre-set to standardize the combination coding method of matrix-related information, plant category information, and plant growth morphology information.

[0068] The facade design prompts are used to guide the generation submodule to generate facade design content that is compatible with substrate type, plant type, and growth form.

[0069] For example, following preset structuring rules for prompts, scattered matrix-related information, plant category information, and plant growth morphology information are linked and integrated to form a logically related multi-dimensional information set. For instance, matrix-related information such as "matrix in brick joint gaps" is used as the appropriate matrix type, plant category information such as "broadleaf herbaceous plants" as the appropriate plant type, and plant growth morphology information such as "growing prostrate along the matrix surface" as the appropriate growth morphology. These are combined and coded according to the preset structuring rules for prompts: "matrix-related information + plant category information + plant growth morphology information," forming facade design prompts that include the appropriate matrix type, plant type, and growth morphology. Thus, by clearly defining the design requirements for matrix type, plant type, and growth morphology through structured facade prompts, the design process of the generated sub-modules has a clear direction, ensuring that the generated building facade design drawings are highly compatible with the substrate to be designed and the ecological conditions of the target area, thereby improving the local adaptability and practicality of the building facade design drawings.

[0070] 2052. Using the constraint submodule, plant growth constraint information is determined based on the masking image and the substrate image to be designed.

[0071] In some embodiments, step 2051 may specifically include the following steps C1 to C5: C1. Perform layer fusion processing on the mask image and the matrix image to be designed to obtain a matrix fusion image with growth area markers.

[0072] Among them, layer fusion processing refers to the process of merging the mask image and the substrate image to be designed according to the preset layer overlay rules, so that the target segmentation area indicated by the mask image is displayed on the substrate image to be designed.

[0073] Among them, the matrix fusion image with growth area markers refers to an image that, after layer fusion processing, simultaneously contains the matrix features of the matrix image to be designed and the target segmentation area markers of the masking image, which can clearly show the location of suitable plant growth areas on the building facade.

[0074] C2. Load the constraint submodule, wherein the constraint submodule includes a semantic segmentation control layer and an image inpainting control layer.

[0075] The semantic segmentation control layer refers to the functional layer in the constraint submodule used to extract semantic features of the matrix region of the image, providing feature support for constructing matrix-related constraint rules.

[0076] Among them, the image inpainting control layer refers to the functional layer in the constraint submodule used to extract features for inpainting the growing region of the image, providing feature support for constructing relevant constraint rules for the growing region.

[0077] C3. Input the matrix fusion image into the constraint submodule, extract semantic features from the matrix fusion image through the semantic segmentation control layer to obtain semantic control features of the matrix region, and extract repair features from the matrix fusion image through the image inpainting control layer to obtain repair control features of the growth region.

[0078] Among them, semantic control features refer to the feature information extracted from the matrix fusion image through the semantic segmentation control layer, which can characterize the attributes of different matrix areas of the building facade.

[0079] Among them, the inpainting control features refer to the feature information extracted from the matrix fusion image through the image inpainting control layer, which can characterize the growth adaptation attributes of the target segmentation region of the building facade.

[0080] C4. Based on the semantic control features and the repair control features, construct regional constraint rules and ecological adaptation constraint rules for plant growth.

[0081] The region constraint rule is used to limit the growth of plants in the target segmentation region indicated by the masking image, and the ecological adaptation constraint rule is used to limit the plant growth morphology to be compatible with the corresponding matrix region.

[0082] C5. Integrate the regional constraint rules and the ecological adaptation constraint rules to form the plant growth constraint information.

[0083] Among them, plant growth constraint information refers to the set of information that can impose multi-dimensional constraints on the plant growth design of generated sub-modules after integrating regional constraint rules and ecological adaptation constraint rules.

[0084] For example, firstly, layer fusion processing is performed on the mask image and the substrate image to be designed, and the target segmentation regions suitable for plant growth are marked on the substrate image to be designed, resulting in a substrate fusion image with growth region markings, providing a unified base image for subsequent feature extraction; for example, the substrate image to be designed is an image of the exterior facade of a brick building, and the mask image is an image marking the target segmentation regions suitable for plant growth, such as the gaps between bricks and the recessed areas of the facade; firstly, the mask image and the substrate image to be designed of the brick facade are layer fused, so that the gaps between bricks and the recessed areas of the facade are clearly marked on the brick facade image, resulting in a substrate fusion image with growth region markings.

[0085] Then, a constraint submodule containing a semantic segmentation control layer and an image inpainting control layer is loaded to provide functional support for feature extraction and constraint rule construction. Subsequently, the matrix fusion image is input into the loaded constraint submodule. The semantic segmentation control layer extracts semantic control features of the matrix region in the image, and the image inpainting control layer extracts inpainting control features of the growth region in the image, providing accurate feature basis for constraint rule construction. Next, based on the extracted semantic control features and inpainting control features, regional constraint rules for limiting plant growth regions and ecological adaptation constraint rules for limiting the adaptation of plant growth morphology to the matrix region are constructed, respectively. For example, a matrix fusion image with growth region identifiers is input into this... The constraint submodule extracts features from the image through a semantic segmentation control layer to obtain semantic control features for different matrix regions, such as the main brick matrix, brick joint gaps, and facade depressions. It also extracts features from the image through an image inpainting control layer to obtain inpainting control features for target segmented regions, such as brick joint gaps and facade depressions. Based on the extracted semantic control features and inpainting control features, the module constructs a region constraint rule: "Plants can only grow in target segmented regions indicated by the masked image, such as brick joint gaps and facade depressions; plants are prohibited from growing in the main brick matrix region." The module also constructs an ecological adaptation constraint rule: "Moss-like plants that attach to the brick joint gaps are suitable for the brick joint gap region; herbaceous plants that grow upright are suitable for the facade depression region."

[0086] Finally, the regional constraint rules and ecological adaptation constraint rules are integrated to form complete plant growth constraint information, providing a clear constraint basis for generating architectural facade design drawings for the sub-modules. For example, the regional constraint rule that "plants can only grow in the target segmented areas indicated by the masked image, such as brick gaps and facade recesses, and plants are prohibited from growing in the main area of ​​the brick matrix" and the ecological adaptation constraint rule that "brick gap areas are suitable for moss-like attached plants, and facade recesses are suitable for herbaceous upright plants" are integrated to form complete plant growth constraint information.

[0087] Therefore, firstly, by merging the growth area identifier of the masked image with the matrix image to be designed through layer fusion processing, a unified matrix fusion image is formed, providing clear and complete basic materials for subsequent feature extraction, thus improving the efficiency and accuracy of feature extraction. Secondly, the constraint submodule extracts features of the matrix region and growth region through semantic segmentation control layer and image inpainting control layer respectively, providing a scientific and precise feature basis for the construction of constraint rules, making the formulation of constraint rules more in line with the actual ecological characteristics of the building facade. Thirdly, by constructing regional constraint rules and ecological adaptation constraint rules respectively, plant growth is constrained from two core dimensions: growth region and growth form, achieving precise and ecological control of plant growth, ensuring that plant growth meets the ecological adaptation requirements of the building facade. Fourthly, by integrating the plant growth constraint information formed by the two types of constraint rules, a complete and clear design constraint basis is provided for the generation submodule, avoiding design results of unreasonable plant growth regions and mismatch between growth forms and matrix regions, thus improving the ecological rationality and practicality of the building facade design drawings.

[0088] 2053. The building facade design drawing is generated by the generation submodule based on the facade design prompt information, the plant growth constraint information and the matrix image to be designed.

[0089] In some embodiments, the generation submodule is trained using the following steps D1 to D6: D1. Using the preset ecological identification module, generate second adaptation description information for the matrix sample image based on the matrix sample image and the ecological description information.

[0090] Among them, the preset ecological identification module is a functional module that can simultaneously process image information (matrix sample map) and text information (ecological description information), and realize cross-modal fusion recognition to output structured ecological description information. Its core function is to identify and extract the core ecological features of the matrix sample map based on the ecological basic data of the target area. Among them, the matrix sample image is a real-world image collected in the target area, containing native plants and corresponding matrix, which is used to provide visual feature learning samples for the training of the generation sub-module. The second adaptation description information is structured text generated by the preset ecological identification module based on the matrix sample map and the ecological description information. It is used to accurately describe the core ecological elements of the matrix sample map, including the adaptation environment characteristics and adaptation plant characteristics of the matrix in the sample map.

[0091] For example, the preset ecological recognition module simultaneously receives matrix sample images and ecological description information. First, it performs visual feature analysis on the matrix sample images to extract core visual elements such as matrix material, plant species, and plant growth morphology. Then, it performs textual semantic analysis on the ecological description information to extract local ecological adaptation patterns of the target area (such as the adaptation relationship between different matrices and plants). Finally, it performs cross-modal fusion of visual elements and local ecological adaptation patterns to generate structured, complete, and highly matched second adaptation description information that is highly compatible with the matrix sample images. This ensures that the second adaptation description information can capture the core ecological features of the matrix sample images and meet the ecological needs of the target area.

[0092] D2. Based on the second adaptation description information and the matrix sample map, construct the first training dataset.

[0093] The first training dataset includes multiple first sample data sets. Each first sample data set includes a matrix sample image and corresponding first sample prompt information and first sample constraint information. The first sample prompt information is determined based on the matrix sample image and the second adaptation description information, and the first sample constraint information is determined based on the second adaptation description information and the matrix sample image. The first sample prompt information is similar to the facade design prompt information, and the first sample constraint information is similar to the plant growth constraint information. The determination methods for the first sample prompt information and the first sample constraint information can be referred to the relevant explanations in steps 2051-2052, and will not be repeated here.

[0094] The first training dataset is a training dataset composed of multiple first sample data. Each first sample data is a combination of "matrix sample map + corresponding first sample prompt information + corresponding first sample constraint information".

[0095] For example, each generated first sample prompt information and first sample constraint information is combined with the corresponding matrix sample map to form a first sample data. By collecting multiple sets of first sample data of different matrix types and different plant varieties, the training set and validation set are divided according to a preset ratio (such as 7:3) and integrated to form a standardized first training dataset, which provides reliable data support for the training and validation of the ecological design module to be trained, and ensures the effectiveness and accuracy of the training.

[0096] D3. Construct the ecological design module to be trained.

[0097] Please refer to the following: Figure 4 , Figure 4This is a schematic diagram of the structural framework of the training ecosystem design module in an embodiment of this application. The training ecosystem design module includes a pre-training basic sub-module and a low-rank matrix sub-module. The pre-training basic sub-module includes a prediction network, a text encoder, an image encoder, and a decoder. The low-rank matrix sub-module is mounted on the attention layer of the prediction network. The text encoder is used to encode the first sample prompt information and output a first text encoding feature. The image encoder is used to encode the matrix sample image and output a first image encoding feature. The prediction network is used to output a first predicted image feature based on the first sample constraint information, the first text encoding feature, and the first image encoding feature. The decoder is used to output a first training design map based on the first predicted image feature.

[0098] Among them, the first image coding feature is a high-dimensional visual feature obtained by the image encoder after encoding the matrix sample image, which is used to characterize the visual attributes of the matrix sample image (such as matrix texture, plant morphology, etc.).

[0099] The first predicted image feature is the image feature predicted and output by the prediction network based on the first sample constraint information, the first text encoding feature, and the first image encoding feature.

[0100] Among them, the first training design map is the design map of the decoder based on the decoding output of the first predicted image features. It is used to compare with the matrix sample map to calculate the first loss value and is the core basis for guiding the adjustment of the parameters of the low-rank matrix submodule.

[0101] For example, a pre-trained basic sub-module is first loaded, which includes a prediction network, a text encoder, an image encoder, and a decoder. Then, a low-rank matrix sub-module is attached to the attention layer of the prediction network to construct a bypass residual structure. This allows the low-rank matrix sub-module to perform targeted learning based on the features of the attention layer, while retaining the original feature processing capabilities of the pre-trained basic sub-module. Finally, a complete, well-defined, and trainable ecological design module is formed, ensuring that the module can accurately learn the ecological adaptation rules of the target region. For example, first, pre-trained basic sub-modules are loaded: the text encoder uses a Transformer encoder to encode the second adaptation description information and output a 768-dimensional first text encoding feature; the image encoder uses a Vision-Transformer to encode the matrix sample image and output a 1024-dimensional first image encoding feature; the prediction network uses a UNet architecture to predict based on the first sample constraint information, the first text encoding feature, and the first image encoding feature, and output a 1024-dimensional first predicted image feature; the decoder uses a VAE decoder to generate a 512×512 pixel first training design map based on the first predicted image feature; then, the low-rank matrix sub-module (with dimensions set to 128×512) is mounted on the attention layer of the UNet architecture, and its output features are residually superimposed with the original output features of the attention layer to form a bypass residual structure, thus completing the construction of the ecological design module to be trained.

[0102] D4. Determine the first loss value of the low-rank matrix submodule based on the image-text pair dataset.

[0103] The first loss value is determined based on the first training design map and the matrix sample map, and is used to guide the parameter adjustment of the low-rank matrix submodule to ensure that the generated first training design map fits the actual ecological and visual characteristics of the matrix sample map.

[0104] For example, multiple first sample data included in the first training dataset are input in batches into the ecological design module to be trained. Each first sample data matrix sample image and its corresponding first sample prompt information are encoded by an image encoder and a text encoder to obtain first image encoding features and first text encoding features, respectively. The prediction network makes predictions based on the first sample constraint information, the first text encoding features, and the first image encoding features, and outputs first predicted image features. The decoder generates a first training design image based on the first predicted image features. An image similarity evaluation algorithm (such as mean squared error MSE, structural similarity SSIM) is used to calculate the difference between each set of first training design images and the corresponding matrix sample images. This difference is quantified as a first loss value, which measures the difference between the two in pixels, structure, and core ecological elements, providing a precise quantitative basis for parameter adjustment of the low-rank matrix submodule.

[0105] D5. Adjust the parameters of the low-rank matrix submodule in the ecosystem design module to be trained according to the first loss value until the training stop condition is met, and then obtain the trained ecosystem design module.

[0106] For example, based on the first loss value calculated in step D4, the backpropagation algorithm is used to keep the parameters of the pre-trained basic sub-modules fixed (to avoid damaging their original basic capabilities and ensure that their text encoding, image encoding, noise reduction, and image generation capabilities are not affected). The parameters of the low-rank matrix sub-modules are mainly iteratively adjusted. After each round of parameter adjustment, the generation effect of the ecological design module to be trained is verified using the image and text dataset validation set, and the first loss value of the validation set is calculated. The iterative adjustment continues until the first loss value of the validation set tends to stabilize and is lower than the preset threshold for a preset number of rounds (e.g., 10 rounds), reaching the condition to stop training. At this time, the trained ecological design module with optimized parameters is obtained, ensuring that the module can accurately generate design maps that fit the ecological characteristics of the target area.

[0107] D6. Based on the model parameters of the trained ecological design module, determine the generation sub-module.

[0108] Once the ecological design module to be trained is complete, the model parameters of the trained ecological design module can be extracted according to its structural framework to construct a generative submodule. This generative submodule includes a prediction network, a text encoder, an image encoder, and a decoder, and can perform similar functions to the prediction network, text encoder, image encoder, and decoder of the trained ecological design module. This allows the generative submodule to accurately learn the design features in the first training dataset and generate building facade design drawings based on facade design prompts, plant growth constraints, and the image of the substrate to be designed.

[0109] For example, the pre-trained basic submodule is trained using the following steps E1 to E4: E1. Obtain the second training dataset for the target region.

[0110] The second training dataset includes multiple second sample data. Each second sample data includes a plant sample image of the target area and corresponding labeled second sample prompt information and second sample constraint information. The plant sample image is a standardized fixed-resolution image. The second sample prompt information is similar to the facade design prompt information, and the second sample constraint information is similar to the plant growth constraint information. The determination method of the second sample prompt information and the second sample constraint information can be referred to the relevant descriptions in steps 2051 to 2052, which will not be repeated here.

[0111] The plant sample image refers to an image collected in the target area that contains native plants and corresponding substrates. It is a standardized fixed-resolution image used to provide visual feature learning samples for a preset image generation model.

[0112] For example, for the target area where the building to be designed is located, real-world images containing native plants and corresponding substrates are collected within that area. The collected images are standardized and uniformly adjusted to a fixed resolution to obtain plant sample images. Simultaneously, second sample prompts and second sample constraints are generated for each plant sample image, ensuring that the format of the second sample prompts matches the facade design prompts and the format of the second sample constraints matches the plant growth constraints. Each plant sample image is combined with its corresponding second sample prompts and second sample constraints to form second sample data. Multiple second sample datasets are integrated to form a second training dataset. Each second sample dataset is a combination of "plant sample image + corresponding second sample prompts + corresponding second sample constraints." Through this pairing of plant sample images and their corresponding second sample prompts, the ecological design module to be trained can learn the adaptation rules between plants and substrates in the target area. Through this pairing of plant sample images and their corresponding second sample constraints, the training of the preset image generation model can learn the spatial constraints between the substrate and plants.

[0113] E2. Obtain the preset image generation model.

[0114] Please refer to the following: Figure 5 , Figure 5 This is a schematic diagram of the structural framework of a preset image generation model provided in this application embodiment. The preset image generation model includes a first image encoding layer, a text encoding layer, a prediction layer, a plant filtering layer, a second image encoding layer, and a decoding layer. The first image encoding layer is used to encode the plant sample image to obtain a second image encoding feature. The text encoding layer is used to encode the second sample prompt information to obtain a second text encoding feature. The plant filtering layer is used to filter out the plants contained in the plant sample image to obtain a filtered image. The second image encoding layer is used to encode the filtered image to obtain a third image encoding feature. The prediction layer is used to output a second predicted image feature based on the second sample constraint information, the second text encoding feature, and the third image encoding feature. The decoding layer is used to generate a second training design map based on the second predicted image feature.

[0115] The second image coding feature refers to the visual features obtained after the first image coding layer encodes the image containing plant samples, which are used to characterize the image attributes of the image containing plant samples.

[0116] Among them, the second text encoding feature refers to the semantic feature obtained after the text encoding layer performs text encoding on the second sample prompt information, which is used to characterize the text attributes of the second sample prompt information.

[0117] The second predicted image feature refers to the predicted image feature output by the prediction layer based on the second sample constraint information, the second text encoding feature, and the third image encoding feature.

[0118] The filtered image refers to the image where the plant filtering layer filters out the plants in the plant sample image and retains only the matrix portion.

[0119] The third image coding feature refers to the visual features obtained after the second image coding layer encodes the filtered image, which are used to characterize the image properties of the matrix.

[0120] The second training design map refers to the design map corresponding to the image containing plant samples generated by the decoding layer based on the features of the second predicted image, which is used to calculate the second loss value during the training process of the preset image generation model.

[0121] The second loss value refers to the difference loss value determined based on the second training design map and the image containing plant samples. It is used to guide the preset image generation model to adjust the parameters of the text encoding layer so that the generated second training design map fits the actual features of the image containing plant samples.

[0122] The third loss value refers to the difference loss value determined based on the second image encoding features and the second text encoding features. It is used to guide the preset image generation model to adjust the parameters of the text encoding layer and strengthen the correlation between the plant sample image and the second sample prompt information.

[0123] Among them, the post-trained image generation model refers to the model that has the ability to learn ecological adaptation of the target region after the preset image generation model is trained on the second training dataset and the text encoding layer parameters are adjusted according to the second loss value and the third loss value.

[0124] For example, a preset image generation model with basic image encoding, text encoding, noise reduction, and image generation capabilities is obtained. This model contains six functional layers: <1> The first image coding layer is responsible for encoding the image containing plant samples and outputting the second image coding features; <2> The text encoding layer is responsible for encoding the second sample prompt information and outputting the second text encoding features; <3> The plant filtering layer is responsible for filtering out the plant parts in the plant sample image to obtain the filtered image. <4> The second image coding layer is responsible for encoding the filtered image and outputting the third image coding features; <5> The prediction layer is responsible for making predictions based on the second sample constraint information, the second text encoding features, and the third image encoding features, and outputting the second predicted image features. <6> The decoding layer is responsible for generating the second training design map based on the features of the second predicted image. For example, a preset image generation model based on the StableDiffusion architecture is obtained. This model includes a first image encoding layer, a text encoding layer, a prediction layer, a plant filtering layer, a second image encoding layer, and a decoding layer. The first image encoding layer uses a VisionTransformer encoder (ViT) to take an image containing plant samples as input and output the second image encoding features. The text encoding layer uses a Transformer encoder structure to encode the second sample prompt information and output the second text encoding features. The plant filtering layer uses a U-Net semantic segmentation model to identify and filter out plant regions in the image containing plant samples, obtaining a filtered image mainly containing the matrix. The second image encoding layer has the same structure as the first image encoding layer, encodes the filtered image, and outputs the third image encoding features. The prediction layer uses an attention mechanism combined with a residual network structure to predict based on the second sample constraint information, the second text encoding features, and the third image encoding features, and outputs the second predicted image features. The decoding layer uses a VAE decoder to generate the second training design map based on the second predicted image features.

[0125] E3. During the training of the preset image generation model using the second training dataset, the parameters of the text encoding layer are adjusted according to the second and third loss values ​​of the text encoding layer to obtain the trained image generation model.

[0126] The second loss value is determined based on the second training design map and the image containing plant samples, and the third loss value is determined based on the second image encoding features and the second text encoding features.

[0127] The third loss value is constructed based on the second image encoding features and the second text encoding features. Since the second text encoding features and the second image encoding features belong to feature information of different data dimensions, in order to construct the second loss, for each second sample data, the second text encoding features obtained by encoding the second sample prompt information through the text encoding layer, and the second image encoding features obtained by encoding the plant sample image through the first image encoding layer, are first processed into a first encoding vector and the second image encoding features are processed into a second encoding vector to map the second text encoding features and the second image encoding features to the same data dimension. Then, the cosine similarity between the first encoding vector and the second encoding vector is calculated as the loss value of the second sample data. The loss values ​​of all second sample data are summed to obtain the second loss value of the text encoding layer.

[0128] For example, a second training dataset is input into a preset image generation model. The various functional layers of the model work collaboratively to complete feature extraction and the generation of the second training design image. During training, a second loss value and a third loss value are calculated: the second loss value is determined based on the difference between the generated second training design image and the original image containing plant samples, used to constrain the visual realism of the second training design image; the third loss value is determined based on the difference between the second image encoding features and the second text encoding features. First, the two types of features are mapped to the same data dimension, and then the similarity is calculated to obtain the loss, used to constrain the association between text and image. Based on the fusion loss of the second and third loss values, the parameters of the text encoding layer are iteratively adjusted until the model converges, resulting in a trained image generation model that ensures the text encoding layer can accurately learn the association between the image containing plant samples and the second sample prompt information. For example, a second training dataset from Shenzhen (500 image-text data pairs) is input into the above preset image generation model. Each layer begins working, and finally, the first image encoding layer outputs the second image encoding features, and the decoding layer outputs the second training design image. During training, the mean squared error (MSE) is used to calculate the difference between the second training design image and the image containing plant samples, yielding the second loss value. The second text encoding features and the second image encoding features are processed into 768-dimensional vectors, and their cosine similarity is calculated. The loss for a single data pair is calculated by subtracting the cosine similarity from 1. The losses for all 500 data pairs are summed to obtain the third loss value. The second and third loss values ​​are then fused with a 1:1 weight to obtain the total loss. Based on the Adam optimizer, the loss is calculated using... The learning rate is iteratively adjusted to adjust the parameters of the Transformer encoder in the text encoding layer. After training for 5 epochs, the model loss value tends to stabilize (converge), resulting in the trained image generation model.

[0129] By constructing a second loss value using the loss between the second training design map and the image containing plant samples, the parameters of the text encoding layer are adjusted so that the text encoding layer can learn the correlation between the image containing plant samples and the second sample prompt information. This allows the text encoding layer to extract the implicit features of the substrate image with plants that are adapted to the ecological environment of the target area, thereby improving the ecological adaptability of the design map generated by the (decoding layer) using the text encoding features.

[0130] E4. Extract the parameters of the prediction layer, the text encoding layer, the second image encoding layer, and the decoding layer of the trained image generation model and fuse them to obtain the pre-trained basic submodule.

[0131] For example, after training is completed, the parameters of the core functional layers in the trained image generation model are extracted to form a pre-trained basic sub-module. The pre-trained basic sub-module includes a prediction network, a text encoder, an image encoder, and a decoder. Specifically, it includes extracting the parameters of the prediction layer as the parameters of the prediction network, extracting the parameters of the text encoding layer as the parameters of the text encoder, extracting the parameters of the second image encoding layer as the parameters of the image encoder, and extracting the parameters of the decoding layer as the parameters of the decoder. This enables the pre-trained basic sub-module to have the core capabilities of text encoding, matrix image encoding, and ecologically adapted building facade design drawing generation. It can be connected to the output of the prompting sub-module and the constraint sub-module to complete the generation of building facade design drawings.

[0132] Therefore, by jointly adjusting the parameters of the text encoding layer based on the second and third loss values, the visual consistency between the second training design drawing and the real plant matrix image is ensured, and the association between text and image is strengthened, enabling the generation submodule to accurately parse the subsequent input facade design prompts and plant growth constraint information.

[0133] Please refer to Figure 6 , Figure 6 This is a structural framework diagram of a building facade design model in an embodiment of this application. The building facade design model includes a preset ecological identification module, a preset masking module, and a preset ecological design module. The preset ecological design module specifically includes a prompting submodule, a constraint submodule, and a generation submodule. Steps 201-205 can be implemented using the building facade design model, such as... Figure 6As shown, the preset ecological identification module takes the target area's substrate image and the ecological description information as input and outputs first adaptation description information (including adaptation environment characteristic information and adaptation plant characteristic information); the preset masking module takes the adaptation environment characteristic information and the substrate image as input and outputs a masking image; the prompting submodule takes the first adaptation description information (including adaptation environment characteristic information and adaptation plant characteristic information) as input and outputs facade design prompt information; the constraint submodule takes the masking image and the substrate image as input and outputs plant growth constraint information; and the generation submodule takes the facade design prompt information, plant growth constraint information, and the substrate image as input and outputs building facade design drawings.

[0134] As can be seen from the above, firstly, by acquiring the ecological description information of the target area and using a preset ecological identification module, the first adaptive description information of the substrate image to be designed is determined based on the ecological description information. This ensures that the first adaptive description information (including adaptive environmental characteristic information and adaptive plant characteristic information) conforms to the environmental and plant characteristics of the target area. Consequently, using the adaptive environmental characteristic information and adaptive plant characteristic information for facade design can automatically analyze the local ecological conditions of the target area for building facade design, improving the ecological adaptability of the building facade design. Secondly, by segmenting the substrate image to be designed based on the adaptive environmental characteristic information to obtain a mask image, and then performing facade design based on the mask image, since the mask image is obtained by segmenting using the adaptive environmental characteristic information, the target segmentation area suitable for plant growth can be clearly identified, providing regional constraints for ecological design. Consequently, when using the mask image for facade design, the plants in the building facade design are distributed in substrate areas that conform to their growth, improving the ecological adaptability of the building facade design. Therefore, the embodiments of this application can improve the matching degree between building facade design drawings and the actual ecological needs of the target area, and improve the ecological adaptability of building facade design schemes.

[0135] Those skilled in the art will understand that all or part of the steps in the above-described method for generating architectural facade design drawings can be accomplished by instructions, or by controlling related hardware through instructions. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0136] Therefore, embodiments of this application provide a computer-readable storage medium storing a plurality of computer programs that can be loaded by a processor to execute any of the architectural facade design drawing generation methods provided in embodiments of this application. For example, the computer program can be loaded by a processor to perform the following steps: The process involves: acquiring an image of the substrate to be designed; acquiring ecological description information of the target area, wherein the ecological description information includes environmental characteristic information of each ecological sample image and plant characteristic information of each plant sample image; using a preset ecological recognition module, identifying the substrate image to be designed and the ecological description information to obtain first adaptation description information of the substrate image to be designed, wherein the first adaptation description information includes adaptation environmental characteristic information and adaptation plant characteristic information of the substrate image to be designed; using a preset masking module, segmenting the substrate image to be designed based on the adaptation environmental characteristic information to obtain a masked image of the substrate image to be designed, wherein the masked image is used to indicate target segmented areas suitable for plant growth; and using a preset ecological design module, performing facade design based on the masked image, the adaptation environmental characteristic information, the adaptation plant characteristic information, and the substrate image to be designed to obtain a building facade design drawing that conforms to the target area.

[0137] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0138] In the above embodiments of the method for generating architectural facade design drawings, computer-readable storage media, and electronic devices, the descriptions of each embodiment have different focuses. For parts not detailed in a particular embodiment, please refer to the relevant descriptions of other embodiments. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes and beneficial effects of the computer-readable storage media, electronic devices, and their corresponding units described above can be referred to the description of the method for generating architectural facade design drawings in the above embodiments, and will not be repeated here.

[0139] The foregoing has provided a detailed description of a method for generating architectural facade design drawings, an electronic device, and a computer-readable storage medium provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for generating architectural facade design drawings, characterized in that, The method includes: Obtain the matrix image to be designed; Obtain ecological description information of the target area, wherein the ecological description information includes environmental characteristic information of each ecological sample image and plant characteristic information of each plant sample image; By using a preset ecological identification module, the design substrate image and the ecological description information are identified to obtain the first adaptation description information of the design substrate image. The first adaptation description information includes the adaptation environment characteristic information and the adaptation plant characteristic information of the design substrate image. The substrate image to be designed is segmented according to the adaptive environment characteristic information by a preset masking module to obtain a masking image of the substrate image to be designed. The masking image is used to indicate the target segmentation area suitable for plant growth. By using a preset ecological design module, facade design is performed based on the masking image, the adaptive environment characteristic information, the adaptive plant characteristic information, and the substrate image to be designed, resulting in a building facade design drawing that conforms to the target area.

2. The method for generating architectural facade design drawings according to claim 1, characterized in that, The preset ecological design module includes a prompt submodule, a constraint submodule, and a generation submodule; The process involves using a preset ecological design module to perform facade design based on the masking image, the adaptive environmental characteristic information, the adaptive plant characteristic information, and the substrate image to be designed, resulting in a building facade design drawing that conforms to the target area. This includes: The prompting submodule performs structured processing based on the adaptive environment characteristic information and the adaptive plant characteristic information to obtain structured facade design prompting information. The constraint submodule determines plant growth constraint information based on the masking image and the substrate image to be designed. The generation submodule generates the building facade design drawing based on the facade design prompts, the plant growth constraint information, and the matrix image to be designed.

3. The method for generating architectural facade design drawings according to claim 2, characterized in that, The structured processing based on the adaptive environment characteristic information and the adaptive plant characteristic information to obtain structured facade design prompt information includes: Extract matrix-related information from the adaptive environment characteristic information, and extract plant category information and plant growth morphology information from the adapted plant characteristic information; According to the preset prompt word structure rules, the matrix-related information, plant category information, and plant growth morphology information are combined and encoded to form facade design prompt information that includes suitable matrix type, suitable plant type, and suitable growth morphology.

4. The method for generating architectural facade design drawings according to claim 2, characterized in that, The step of determining plant growth constraint information through the constraint submodule based on the masking image and the substrate image to be designed includes: The mask image and the matrix image to be designed are layer fusion processed to obtain a matrix fusion image with growth area markers; Load the constraint submodule, wherein the constraint submodule includes a semantic segmentation control layer and an image inpainting control layer; The matrix fusion image is input into the constraint submodule. The semantic segmentation control layer extracts semantic features from the matrix fusion image to obtain semantic control features of the matrix region. The image inpainting control layer extracts inpainting features from the matrix fusion image to obtain inpainting control features of the growth region. Based on the semantic control features and the repair control features, regional constraint rules and ecological adaptation constraint rules for plant growth are constructed. The regional constraint rules are used to limit the growth of plants in the target segmentation region indicated by the masking image, and the ecological adaptation constraint rules are used to limit the plant growth morphology to be adapted to the corresponding matrix region. The regional constraint rules and the ecological adaptation constraint rules are integrated to form the plant growth constraint information.

5. The method for generating architectural facade design drawings according to claim 2, characterized in that, The generation submodule is trained in the following manner: The preset ecological identification module generates second adaptation description information for the matrix sample image based on the matrix sample image and the ecological description information. Based on the second adaptation description information and the matrix sample map, a first training dataset is constructed, wherein the first training dataset includes multiple first sample data, each first sample data includes a matrix sample map and corresponding first sample prompt information and first sample constraint information, the first sample prompt information is determined based on the matrix sample map and the second adaptation description information, and the first sample constraint information is based on the second adaptation description information and the matrix sample map; An ecosystem design module to be trained is constructed, comprising a pre-training basic sub-module and a low-rank matrix sub-module. The pre-training basic sub-module includes a prediction network, a text encoder, an image encoder, and a decoder. The low-rank matrix sub-module is mounted on the attention layer of the prediction network. The text encoder encodes the first sample prompt information and outputs a first text encoding feature. The image encoder encodes the matrix sample image and outputs a first image encoding feature. The prediction network outputs a first predicted image feature based on the first sample constraint information, the first text encoding feature, and the first image encoding feature. The decoder outputs a first training design map based on the first predicted image feature. Based on the first training dataset, a first loss value for the low-rank matrix submodule is determined, wherein the first loss value is determined based on the first training design graph and the matrix sample graph; The parameters of the low-rank matrix submodule in the ecological design module to be trained are adjusted according to the first loss value until the training stop condition is met, and the trained ecological design module is obtained. The generation submodule is determined based on the model parameters of the trained ecological design module.

6. The method for generating architectural facade design drawings according to claim 5, characterized in that, The pre-trained basic submodule is trained in the following manner: Obtain a second training dataset for the target region, wherein the second training dataset includes multiple second sample data, and each second sample data includes a plant sample map of the target region and corresponding labeled second sample prompt information and second sample constraint information; A preset image generation model is obtained, wherein the preset image generation model includes a first image encoding layer, a text encoding layer, a prediction layer, a plant filtering layer, a second image encoding layer, and a decoding layer. The first image encoding layer is used to encode the plant sample image to obtain a second image encoding feature. The text encoding layer is used to encode the second sample prompt information to obtain a second text encoding feature. The plant filtering layer is used to filter out the plants contained in the plant sample image to obtain a filtered image. The second image encoding layer is used to encode the filtered image to obtain a third image encoding feature. The prediction layer is used to output a second predicted image feature based on the second sample constraint information, the second text encoding feature, and the third image encoding feature. The decoding layer is used to generate a second training design image based on the second predicted image feature. During the training of the preset image generation model using the second training dataset, the parameters of the text encoding layer are adjusted according to the second loss value and the third loss value of the text encoding layer to obtain the trained image generation model. The second loss value is determined based on the second training design map and the image containing plant samples, and the third loss value is determined based on the second image encoding feature and the second text encoding feature. The parameters of the prediction layer, the text encoding layer, the second image encoding layer, and the decoding layer of the trained image generation model are extracted and fused to obtain the pre-trained basic submodule.

7. The method for generating architectural facade design drawings according to claim 1, characterized in that, The adaptation environment characteristic information is identified in the following way: The matrix image to be designed is embedded to obtain a first image embedding vector; Based on the first image embedding vector and the image embedding vectors of each of the ecological sample images, a target ecological sample image that matches the matrix of the matrix image to be designed is determined; Environmental characteristic information of the target ecological sample image is obtained based on the ecological description information, and is used as the adaptive environmental characteristic information.

8. The method for generating architectural facade design drawings according to claim 7, characterized in that, The adaptive plant characteristic information is identified in the following way: Obtain the second image embedding vector of the target ecological sample image; Based on the second image embedding vector and the image embedding vectors of each plant sample image, a target plant sample image that matches the plant in the target ecological sample image is determined; Based on the ecological description information, plant characteristic information of the target plant sample image is obtained as the adaptive plant characteristic information.

9. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, and the processor executes the method for generating architectural facade design drawings as described in any one of claims 1 to 8 when it invokes the computer program in the memory.

10. A computer-readable storage medium, characterized in that, It stores a computer program, which is loaded by a processor to execute the method for generating architectural facade design drawings according to any one of claims 1 to 8.

Images