Extraterrestrial building image generation method based on generative artificial intelligence

Abstract

The invention discloses an extraterrestrial building image generation method based on generative artificial intelligence, and relates to the field of extraterrestrial building image generation. The objective of the invention is to solve the problems of low generation efficiency and low accuracy of an existing extraterrestrial building image generation method. The method comprises the steps of obtaining a description text of a to-be-generated extraterrestrial building image, and inputting the description text of the to-be-generated extraterrestrial building image into an extraterrestrial building image generation model to obtain an extraterrestrial building image; the description text of the to-be-generated extraterrestrial building image comprises a building shape, a building material, a building structure, a building height and a building skin; the extraterrestrial building image generation model is obtained by the following steps: obtaining an extraterrestrial building picture, preprocessing the extraterrestrial building picture, and establishing an extraterrestrial building material data set by using the preprocessed extraterrestrial building picture; and obtaining an extraterrestrial building image generation model by using the extraterrestrial building material data set. The method is used for generating the extraterrestrial building image.

Description

Technical Field

[0001] The present invention relates to the field of extraterrestrial building image generation, and in particular to an extraterrestrial building image generation method based on generative artificial intelligence. Background Art

[0002] With the development of human science and technology, extraterrestrial exploration and immigration have become possible, especially the exploration of extraterrestrial bodies such as the moon and Mars has become the focus of international cooperation. Extraterrestrial architecture is the outpost and habitat for human exploration of the universe, and has very important scientific exploration significance and development strategic significance. Extraterrestrial architectural design is a systematic and complex issue. In the process of extraterrestrial architectural design, we must not only consider the safety of survival on alien planets, but also the rationality of how humans can live efficiently and comfortably under the constraints of complex environments. Therefore, it is very important to obtain extraterrestrial architectural images that meet the design requirements in the process of extraterrestrial architectural design.

[0003] At present, the generation of images of extraterrestrial buildings relies on the manual design of the appearance of extraterrestrial buildings. Designers conceive the appearance of extraterrestrial buildings based on their own experience and theoretical speculation, and present and disseminate them in the form of images. However, it often takes a long time to conceive and discuss the design and generation of extraterrestrial building images manually, and it is not possible to quickly generate a variety of extraterrestrial building images, which makes it difficult to choose for subsequent actual design of extraterrestrial buildings. In addition, different designers have different understandings of design requirements, which leads to low accuracy of extraterrestrial building images drawn by designers. Therefore, the current method of generating extraterrestrial building images still has the problems of low generation efficiency and accuracy. Summary of the invention

[0004] The purpose of the present invention is to solve the problems of low generation efficiency and accuracy in the existing extraterrestrial building image generation methods, and to propose an extraterrestrial building image generation method based on generative artificial intelligence.

[0005] The method for generating images of extraterrestrial buildings based on generative artificial intelligence is as follows:

[0006] Obtaining a description text of the extraterrestrial building image to be generated, inputting the description text of the extraterrestrial building image to be generated into an extraterrestrial building image generation model, and obtaining the extraterrestrial building image;

[0007] The description text of the extraterrestrial building image to be generated includes: building shape, building materials, building structure, building height, and building surface;

[0008] The extraterrestrial building image generation model is obtained by the following method:

[0009] Step 1: Obtain extraterrestrial building pictures, pre-process the extraterrestrial building pictures, and use the pre-processed extraterrestrial building pictures to establish an extraterrestrial building material data set;

[0010] Step 2: Use the extraterrestrial building material dataset to obtain an extraterrestrial building image generation model.

[0011] Furthermore, in the step 1, the extraterrestrial building pictures are obtained, and the extraterrestrial building pictures are preprocessed, and the extraterrestrial building material data set is established using the preprocessed extraterrestrial building pictures, specifically:

[0012] Step 11: Obtain extraterrestrial building images, unify the resolution and size of the extraterrestrial building images, and obtain pre-processed extraterrestrial building images;

[0013] Step 1 and 2, obtaining keyword texts of the preprocessed extraterrestrial building images;

[0014] The keyword text stores extraterrestrial building features, including: building shape, building materials, building structure, building height, and building surface;

[0015] Step 13: Check whether the content in the keyword text conforms to the architectural features of the corresponding off-site architectural picture, delete the keyword text content that does not conform to the architectural features, and then add trigger words to the keyword text;

[0016] Step 14: Use extraterrestrial building images and corresponding keyword texts to form an extraterrestrial building material dataset.

[0017] Furthermore, the step 2 of using the extraterrestrial building material dataset to obtain an extraterrestrial building image generation model is specifically as follows:

[0018] Step 21: Use the extraterrestrial building material dataset as a training set, use the training set to train the low-rank matrix model, and obtain a trained low-rank matrix model;

[0019] Step 22: Load the trained low-rank matrix model into the initial benchmark model to obtain the final benchmark model;

[0020] The initial benchmark model is Stable Difussion or ChilloutMix;

[0021] Step 23: Test the final benchmark model. If the final benchmark model meets the preset effect, the final benchmark model will be used as the extraterrestrial building image generation model; otherwise, adjust the training parameters and sampling parameters of the low-rank matrix model and return to step 21.

[0022] Furthermore, in step 22, the trained low-rank matrix model is loaded into the initial benchmark model to obtain the final benchmark model, as follows:

[0023] W=W0+L0

[0024] Among them, W0 is the initial benchmark model parameter, L0 is the parameter of the low-rank matrix model, and W is the final benchmark model parameter.

[0025] Furthermore, the preset effects in steps two and three are as follows:

[0026] In the final benchmark model, more than 50% of the extraterrestrial building images have extraterrestrial building features that match the extraterrestrial building features in the keyword text input during the test without overfitting;

[0027] The overfitting phenomenon is that the extraterrestrial building features of more than 50% of the extraterrestrial building images in the generated extraterrestrial building pictures are completely identical to at least two extraterrestrial building features of the extraterrestrial building images in the extraterrestrial building material dataset.

[0028] The specific process of the extraterrestrial building image generation method based on generative artificial intelligence is as follows:

[0029] Step 1, obtaining an artificial design sketch of an extraterrestrial building, converting the artificial design sketch of the extraterrestrial building into a black and white line draft, and obtaining the artificial design sketch of the extraterrestrial building after the black and white line draft;

[0030] Step 2, inputting the artificial design sketch of the extraterrestrial building after the black and white line draft into the auxiliary neural network model to obtain the outline position information of the extraterrestrial building in the artificial design sketch of the extraterrestrial building after the black and white line draft;

[0031] The auxiliary neural network model is a ControlNet model;

[0032] Step 3, constructing a description text of the extraterrestrial building image to be generated, inputting the description text of the extraterrestrial building image to be generated and the extraterrestrial building outline position information into the extraterrestrial building image generation model to obtain the extraterrestrial building image;

[0033] The description text of the extraterrestrial building image to be generated includes: building shape, building materials, building structure, building height, and building surface;

[0034] The extraterrestrial building image generation model is obtained by the following method:

[0035] S1. Acquire extraterrestrial building pictures, preprocess the extraterrestrial building pictures, and use the preprocessed extraterrestrial building pictures to establish an extraterrestrial building material data set;

[0036] S2. Using the extraterrestrial building material dataset, obtain an extraterrestrial building image generation model.

[0037] Furthermore, the step of obtaining the extraterrestrial building pictures in S1, preprocessing the extraterrestrial building pictures, and using the preprocessed extraterrestrial building pictures to establish an extraterrestrial building material data set is specifically as follows:

[0038] S101, obtaining different extraterrestrial building pictures, and unifying the resolution and size of the extraterrestrial building pictures to obtain pre-processed extraterrestrial building pictures;

[0039] S102, obtaining keyword text of the preprocessed extraterrestrial building picture;

[0040] The keyword text stores extraterrestrial building features, including: building shape, building materials, building structure, building height, and building surface;

[0041] S103, checking whether the content in the keyword text conforms to the architectural features of the corresponding off-site architectural picture, deleting the keyword text content that does not conform to the architectural features, and then adding trigger words to the keyword text;

[0042] S104. Using the extraterrestrial building pictures and corresponding keyword texts, an extraterrestrial building material data set is formed.

[0043] Furthermore, the method of obtaining an extraterrestrial building image generation model using an extraterrestrial building material dataset in S2 is specifically as follows:

[0044] S201, using the extraterrestrial building material dataset as a training set, and using the training set to train a low-rank matrix model to obtain a trained low-rank matrix model;

[0045] S202, loading the trained low-rank matrix model into the initial benchmark model to obtain a final benchmark model;

[0046] The initial benchmark model is Stable Difussion or ChilloutMix;

[0047] S203, testing the final benchmark model. If the final benchmark model meets the preset effect, the final benchmark model is used as the extraterrestrial building image generation model; otherwise, the training parameters and sampling parameters of the low-rank matrix model are adjusted and the process returns to S201.

[0048] Furthermore, in S02, the trained low-rank matrix model is loaded into the initial benchmark model to obtain the final benchmark model, as follows:

[0049] W=W0+L0

[0050] Among them, W0 is the initial benchmark model parameter, L0 is the parameter of the low-rank matrix model, and W is the final benchmark model parameter.

[0051] Furthermore, the preset effect in S203 is as follows:

[0052] More than 50% of the extraterrestrial building images generated by the final benchmark model have extraterrestrial building features that match the extraterrestrial building features in the keyword text input during the test and no overfitting occurs;

[0053] The overfitting means that the extraterrestrial building features of more than 50% of the extraterrestrial building images in the generated extraterrestrial building pictures are completely identical to at least two extraterrestrial building features of the extraterrestrial building images in the extraterrestrial building material dataset.

[0054] The beneficial effects of the present invention are:

[0055] The present invention introduces a generative artificial intelligence method into an extraterrestrial building image generation method, and solves the problem that the generation result does not meet the requirements due to the lack of training set content by training a low-rank matrix model suitable for the design of extraterrestrial building schemes. The present invention introduces an auxiliary neural network architecture model that can generate extraterrestrial building images that meet the requirements of the manual draft description through an artificial design draft, thereby improving the generation efficiency of extraterrestrial building images, thereby assisting extraterrestrial building designers to improve the design. The present invention uses a pre-trained low-rank matrix model and an auxiliary neural network architecture model loaded with a hand-drawn sketch to intervene in the generation process of a generative artificial intelligence large model, solving the problem that the use effect of generative artificial intelligence is completely unsatisfactory when applied to the design scene of extraterrestrial buildings due to the lack of training sets for extraterrestrial buildings. The present invention trains a low-rank matrix model suitable for extraterrestrial buildings, filling the technical gap in generating extraterrestrial building images. The low-rank matrix model can efficiently fine-tune the large model through a small amount of data, and can generate extraterrestrial building images with specific forms and styles for image generation tasks. The low-rank matrix model trained by the present invention can make the generated extraterrestrial building images meet the characteristics of extraterrestrial buildings, thereby improving the accuracy of extraterrestrial building image generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a flow chart of the present invention;

[0057] Figure 2 Sketching artificial designs for extraterrestrial architecture;

[0058] Figure 3 A diagram showing the results. DETAILED DESCRIPTION

[0059] Specific implementation method 1: Figure 1 As shown, the specific process of the extraterrestrial building image generation method based on generative artificial intelligence in this embodiment is as follows:

[0060] Obtaining a description text of the extraterrestrial building image to be generated, inputting the description text of the extraterrestrial building image to be generated into an extraterrestrial building image generation model, and obtaining the extraterrestrial building image;

[0061] The description text of the extraterrestrial building image to be generated includes: building shape, building materials, building structure, building height, and building surface;

[0062] The extraterrestrial building image generation model is obtained by the following method:

[0063] Step 1: Obtain extraterrestrial building images, preprocess the extraterrestrial building images, and use the preprocessed extraterrestrial building images to establish an extraterrestrial building material dataset, specifically:

[0064] Step 11: Obtain extraterrestrial building images of different styles, unify the resolution and size of the extraterrestrial building images, and obtain pre-processed extraterrestrial building images:

[0065] In this step, obtain 50-100 pictures of extraterrestrial buildings of different styles; the pictures should be clear and distinctive, avoid cluttered elements such as logos and banners, and reduce the number of pictures with high repetition and similarity. Unify the resolution of extraterrestrial building pictures and crop them to the same size. In actual application, it is recommended to unify the resolution of extraterrestrial building pictures to a multiple of 64 and crop the extraterrestrial building pictures to 512*512, 512*768 or 768*768;

[0066] Step 1 and 2, obtaining keyword texts of the preprocessed extraterrestrial building images;

[0067] The keyword text stores extraterrestrial building features, including building shape, building materials, building structure, building height, building surface, etc.;

[0068] The keyword text language is English;

[0069] The content stored in the keyword text may be a word or a sentence;

[0070] The keyword text is generated manually or by an existing natural language processing package;

[0071] Step 13: Check whether the content in the keyword text meets the architectural features of the corresponding foreign architectural pictures, delete the text content that does not meet the features, and then add trigger words to the keyword text;

[0072] The trigger words are used as instructions during the training process to call the low-rank matrix model to control the generation process of the extraterrestrial building plan. Once the trigger words are entered, the low-rank matrix model intervenes in the generation process of the generative artificial intelligence large model, so that the generated images all meet the characteristics of extraterrestrial buildings. If no trigger words are entered, even if the low-rank matrix model is loaded, the model will not interfere with the generation process of the generative artificial intelligence large model.

[0073] For example, trigger words can be "space", "extraterrestrial architecture" and other words related to extraterrestrial architectural design;

[0074] Step 14: Package the extraterrestrial building images and corresponding keyword texts to form an extraterrestrial building material dataset.

[0075] Step 2: Use the extraterrestrial building material dataset to obtain an extraterrestrial building image generation model, specifically:

[0076] Step 21: Use the extraterrestrial architectural material dataset as a model training set, and use the training set to train a low-rank matrix model. When the loss function of the training low-rank matrix model is in the range of 0.12-0.15, a trained low-rank matrix model is obtained;

[0077] During the training process, adjust the training parameters and sampling parameters such as the number of learning times, number of cycles, number of parallelism, and learning accuracy to keep the loss function within the range of 0.12-0.15;

[0078] Step 22: Load the trained low-rank matrix model into the initial benchmark model to obtain the final benchmark model;

[0079] The process of the initial benchmark model calling the trained low-rank matrix model can be regarded as a linear relationship, which can be expressed by the data formula as follows. The parameters of the final benchmark model are:

[0080] W=W0+L0

[0081] Among them, W0 is the initial benchmark model parameter, L0 is the parameter of the low-rank matrix model, and W represents the final benchmark model parameter after the low-rank matrix model is introduced;

[0082] The initial benchmark model is a large model such as Stable Difussion or ChilloutMix, or a self-trained large model;

[0083] Since the text-generated image artificial intelligence model can adjust the image output results based on text descriptions, which is in line with the usage habits of extraterrestrial architectural designers, the present invention uses the text-generated image generative artificial intelligence large model as the benchmark model and loads a pre-trained low-rank matrix model into the benchmark model.

[0084] Step 23: Test the final benchmark model, and use the image generation result of the final benchmark model to reflect the stylized regulation ability of the low-rank matrix model on the benchmark model. During the test, input the prompt text and trigger words with exactly the same content, generate about 50 images for detection, and observe the generation effect of the extraterrestrial building image. If it meets the preset effect, the final benchmark model is used as the extraterrestrial building image generation model; otherwise, adjust the training parameters and sampling parameters of the low-rank matrix model and return to step 21;

[0085] The meaning of meeting the preset effect is that more than 50% of the features of the generated images meet the features of the input extraterrestrial buildings, including building shape, building materials, building structure, building height, building surface, etc., and there is no overfitting phenomenon;

[0086] The overfitting phenomenon is: more than 50% of the generated image content is completely consistent with the image content in the extraterrestrial building material data set in two or more features of building shape, building material, building structure, building height, and building surface;

[0087] Poor results mean that the generated image is too similar to the image content in the material set or does not conform to the characteristics of extraterrestrial architecture;

[0088] If all the generated images do not meet the characteristics of extraterrestrial architecture, underfitting occurs. Models with overfitting and underfitting cannot pass the test and need to be abandoned and returned to step 21 to adjust parameters and retrain. Overfitting models need to reduce the number of training times and learning accuracy, while underfitting models need to increase the number of training times and learning accuracy.

[0089] The present invention proposes two application scenarios of generative artificial intelligence in the design process of extraterrestrial architectural schemes. One is to give full play to the creativity of generative artificial intelligence, generate a large number of images for heuristic design, and provide a basis for the scheme conception of extraterrestrial architectural designers. The other is to optimize the design based on the hand-drawn sketches of designers, and use generative artificial intelligence technology to modify the hand-drawn sketches to form a complete extraterrestrial architectural design scheme image, thereby assisting designers in designing. The specific process is shown in the second specific implementation method.

[0090] Specific implementation method 2: A method for generating an extraterrestrial building image based on generative artificial intelligence, comprising the following steps:

[0091] Step 1, obtain the artificial design sketch of the extraterrestrial building, and convert the artificial design sketch of the extraterrestrial building into black and white line draft to obtain the artificial design sketch of the extraterrestrial building after black and white line draft, such as Figure 2 As shown;

[0092] Step 2, inputting the artificial design sketch of the extraterrestrial building after the black and white line draft into the auxiliary neural network model to obtain the outline position information of the extraterrestrial building in the artificial design sketch of the extraterrestrial building after the black and white line draft;

[0093] The auxiliary neural network model is a ControlNet model;

[0094] The parameters of the auxiliary neural network model are the parameters of the encoding layer of the baseline model;

[0095] Step 3, construct the description text of the extraterrestrial building image to be generated, input the description text of the extraterrestrial building image to be generated and the extraterrestrial building outline position information into the extraterrestrial building image generation model, and obtain the extraterrestrial building image, such as Figure 3 As shown;

[0096] The description text of the extraterrestrial building image to be generated includes: building shape, building materials, building structure, building height, and building surface;

[0097] The extraterrestrial building image generation model is obtained by the following method:

[0098] S1. Obtain extraterrestrial building images, preprocess the extraterrestrial building images, and use the preprocessed extraterrestrial building images to establish an extraterrestrial building material data set, specifically:

[0099] S101, acquiring extraterrestrial building pictures of different styles, and unifying the resolution and size of the extraterrestrial building pictures to obtain pre-processed extraterrestrial building pictures;

[0100] S102, obtaining keyword text of the preprocessed extraterrestrial building picture;

[0101] The keyword text stores extraterrestrial building features, including building shape, building materials, building structure, building height, building surface, etc.;

[0102] The keyword text language is English;

[0103] The content stored in the keyword text may be a word or a sentence;

[0104] The keyword text is generated manually or by an existing natural language processing package;

[0105] S103, checking whether the content in the keyword text conforms to the architectural features of the corresponding foreign architectural picture, deleting the text content that does not conform to the features, and then adding trigger words to the keyword text;

[0106] For example, trigger words can be "space", "extraterrestrial architecture" and other words related to extraterrestrial architectural design;

[0107] S104, packaging the extraterrestrial building images and the corresponding keyword texts to form an extraterrestrial building material data set.

[0108] S2. Using the extraterrestrial building material dataset to obtain an extraterrestrial building image generation model, specifically:

[0109] S201, using the extraterrestrial building material data set as a model training set, using the training set to train a low-rank matrix model, and when the loss function of the training low-rank matrix model is in the range of 0.12-0.15, a trained low-rank matrix model is obtained;

[0110] During the training process, adjust the training parameters and sampling parameters such as the number of learning times, number of cycles, number of parallelism, and learning accuracy to keep the loss function within the range of 0.12-0.15;

[0111] S202, loading the trained low-rank matrix model into the initial benchmark model;

[0112] The process of the initial benchmark model calling the trained low-rank matrix model can be regarded as a linear relationship, which can be expressed by the data formula as follows. The parameters of the final benchmark model are:

[0113] W=W0+L0

[0114] Among them, W0 is the initial benchmark model parameter, L0 is the parameter of the low-rank matrix model, and W represents the final benchmark model parameter after the low-rank matrix model is introduced;

[0115] The initial benchmark model is a large model such as Stable Difussion or ChilloutMix, or a self-trained large model;

[0116] S203, using the test set to test the benchmark model after loading the trained low-rank matrix model, and using the image generation result of the benchmark model to reflect the stylized regulation ability of the low-rank matrix model on the benchmark model. During the test, input the prompt word text and trigger word with exactly the same content, generate about 50 images for detection, and observe the generation effect of the extraterrestrial building image. If the effect is good, the trained benchmark model after loading the trained low-rank matrix model is used as the extraterrestrial building image generation model; otherwise, adjust the training parameters and sampling parameters of the low-rank matrix model and return to S201;

[0117] Good results mean that more than 50% of the features of the generated images meet the features of the extraterrestrial buildings mentioned in steps 1 and 2, including building shape, building materials, building structure, building height, building surface, etc., and there is no overfitting phenomenon;

[0118] The overfitting phenomenon is: more than 50% of the generated image content is completely consistent with the image content in the material collection in two or more aspects of building shape, building materials, building structure, building height, and building surface;

[0119] Poor results mean that the generated image is too similar to the image content in the material set or does not conform to the characteristics of extraterrestrial architecture;

[0120] If all the generated images fail to meet the characteristics of extraterrestrial architecture, underfitting occurs. Models with overfitting and underfitting cannot pass the test and need to be abandoned and return to S201 to adjust parameters and retrain. Overfitting models need to reduce the number of training times and learning accuracy, while underfitting models need to increase the number of training times and learning accuracy.

Claims

1. A method for generating images of extraterrestrial buildings based on generative artificial intelligence, characterized by: The specific process of the method is: Obtaining a description text of the extraterrestrial building image to be generated, inputting the description text of the extraterrestrial building image to be generated into an extraterrestrial building image generation model, and obtaining the extraterrestrial building image; The description text of the extraterrestrial building image to be generated includes: building shape, building materials, building structure, building height, and building surface; The extraterrestrial building image generation model is obtained by the following method: Step 1: Obtain extraterrestrial building pictures, pre-process the extraterrestrial building pictures, and use the pre-processed extraterrestrial building pictures to establish an extraterrestrial building material data set; Step 2: Use the extraterrestrial building material dataset to obtain an extraterrestrial building image generation model.

2. The method for generating images of extraterrestrial buildings based on generative artificial intelligence according to claim 1, characterized in that: The step 1 of obtaining the extraterrestrial building pictures, preprocessing the extraterrestrial building pictures, and using the preprocessed extraterrestrial building pictures to establish an extraterrestrial building material data set is specifically as follows: Step 11: Obtain extraterrestrial building images, unify the resolution and size of the extraterrestrial building images, and obtain pre-processed extraterrestrial building images; Step 1 and 2, obtaining keyword texts of the preprocessed extraterrestrial building images; The keyword text stores extraterrestrial building features, including: building shape, building materials, building structure, building height, and building surface; Step 13: Check whether the content in the keyword text conforms to the architectural features of the corresponding off-site architectural picture, delete the keyword text content that does not conform to the architectural features, and then add trigger words to the keyword text; Step 14: Use extraterrestrial building images and corresponding keyword texts to form an extraterrestrial building material dataset.

3. The method for generating images of extraterrestrial buildings based on generative artificial intelligence according to claim 2, characterized in that: The step 2 of using the extraterrestrial building material dataset to obtain an extraterrestrial building image generation model is specifically as follows: Step 21: Use the extraterrestrial building material dataset as a training set, use the training set to train the low-rank matrix model, and obtain a trained low-rank matrix model; Step 22: Load the trained low-rank matrix model into the initial benchmark model to obtain the final benchmark model; The initial benchmark model is Stable Difussion or ChilloutMix; Step 23: Test the final benchmark model. If the final benchmark model meets the preset effect, the final benchmark model will be used as the extraterrestrial building image generation model; otherwise, adjust the training parameters and sampling parameters of the low-rank matrix model and return to step 21.

4. The method for generating images of extraterrestrial buildings based on generative artificial intelligence according to claim 3, characterized in that: In step 22, the trained low-rank matrix model is loaded into the initial benchmark model to obtain the final benchmark model, as follows: W=W0+L0 Among them, W0 is the initial benchmark model parameter, L0 is the parameter of the low-rank matrix model, and W is the final benchmark model parameter.

5. The method for generating images of extraterrestrial buildings based on generative artificial intelligence according to claim 4, characterized in that: The preset effects in steps 2 and 3 are as follows: In the final benchmark model, more than 50% of the extraterrestrial building images have extraterrestrial building features that match the extraterrestrial building features in the keyword text input during the test without overfitting; The overfitting phenomenon is that the extraterrestrial building features of more than 50% of the extraterrestrial building images in the generated extraterrestrial building pictures are completely identical to at least two extraterrestrial building features of the extraterrestrial building images in the extraterrestrial building material dataset.

6. A method for generating images of extraterrestrial buildings based on generative artificial intelligence, characterized in that: The specific process of the method is: Step 1, obtaining an artificial design sketch of an extraterrestrial building, converting the artificial design sketch of the extraterrestrial building into a black and white line draft, and obtaining the artificial design sketch of the extraterrestrial building after the black and white line draft; Step 2, inputting the artificial design sketch of the extraterrestrial building after the black and white line draft into the auxiliary neural network model to obtain the outline position information of the extraterrestrial building in the artificial design sketch of the extraterrestrial building after the black and white line draft; The auxiliary neural network model is a ControlNet model; Step 3, constructing a description text of the extraterrestrial building image to be generated, inputting the description text of the extraterrestrial building image to be generated and the extraterrestrial building outline position information into the extraterrestrial building image generation model to obtain the extraterrestrial building image; The description text of the extraterrestrial building image to be generated includes: building shape, building materials, building structure, building height, and building surface; The extraterrestrial building image generation model is obtained by the following method: S1. Acquire extraterrestrial building pictures, preprocess the extraterrestrial building pictures, and use the preprocessed extraterrestrial building pictures to establish an extraterrestrial building material data set; S2. Using the extraterrestrial building material dataset, obtain an extraterrestrial building image generation model.

7. The method for generating images of extraterrestrial buildings based on generative artificial intelligence according to claim 6, characterized in that: The step S1 of obtaining extraterrestrial building pictures, preprocessing the extraterrestrial building pictures, and using the preprocessed extraterrestrial building pictures to establish an extraterrestrial building material data set is specifically as follows: S101, obtaining different extraterrestrial building pictures, and unifying the resolution and size of the extraterrestrial building pictures to obtain pre-processed extraterrestrial building pictures; S102, obtaining keyword text of the preprocessed extraterrestrial building picture; The keyword text stores extraterrestrial building features, including: building shape, building materials, building structure, building height, and building surface; S103, checking whether the content in the keyword text conforms to the architectural features of the corresponding off-site architectural picture, deleting the keyword text content that does not conform to the architectural features, and then adding trigger words to the keyword text; S104. Using the extraterrestrial building pictures and corresponding keyword texts, an extraterrestrial building material data set is formed.

8. The method for generating images of extraterrestrial buildings based on generative artificial intelligence according to claim 7, characterized in that: The method of obtaining an extraterrestrial building image generation model by using an extraterrestrial building material dataset in S2 is specifically as follows: S201, using the extraterrestrial building material dataset as a training set, and using the training set to train a low-rank matrix model to obtain a trained low-rank matrix model; S202, loading the trained low-rank matrix model into the initial benchmark model to obtain a final benchmark model; The initial benchmark model is Stable Difussion or ChilloutMix; S203, testing the final benchmark model. If the final benchmark model meets the preset effect, the final benchmark model is used as the extraterrestrial building image generation model; otherwise, the training parameters and sampling parameters of the low-rank matrix model are adjusted and the process returns to S201.

9. The method for generating images of extraterrestrial buildings based on generative artificial intelligence according to claim 8, characterized in that: In S02, the trained low-rank matrix model is loaded into the initial benchmark model to obtain the final benchmark model, as follows: W=W0+L0 Among them, W0 is the initial benchmark model parameter, L0 is the parameter of the low-rank matrix model, and W is the final benchmark model parameter.

10. The method for generating images of extraterrestrial buildings based on generative artificial intelligence according to claim 9, characterized in that: The preset effect in S203 is as follows: More than 50% of the extraterrestrial building images generated by the final benchmark model have extraterrestrial building features that match the extraterrestrial building features in the keyword text input during the test and no overfitting occurs; The overfitting means that the extraterrestrial building features of more than 50% of the extraterrestrial building images in the generated extraterrestrial building pictures are completely identical to at least two extraterrestrial building features of the extraterrestrial building images in the extraterrestrial building material dataset.

Images