Architectural drawing recognition method based on AI building drawing recognition

Abstract

The invention relates to the technical field of buildings, in particular to an architectural drawing recognition method based on AI architectural drawing recognition, which comprises the following steps of: training a deformation-based auto-encoder, and mapping a CAD (Computer Aided Design) mark to a potential space; inputting image denoising potential CAD codes by applying potential diffusion deformation and conditions; enhancing the potential diffusion model through additional constraint and regularization; rLHF is introduced, the effectiveness of denoising potential codes is improved through CAD code inspection, and the classified codes are valid or invalid; a diffusion model is finely adjusted through direct preference optimization, and a classification set is used; extracting a classification set and adding an identifier for retrieving information and constructing a metagraph; gAN is used for carrying out adversarial training on potential space representation, and the authenticity and diversity of drawings are improved; a cross-domain transfer learning strategy is adopted to realize identification and conversion of different styles of architectural drawings; the scheme provided by the design shows high flexibility and expandability, and the method for iteratively merging the meta-graphs can efficiently integrate building drawing information to form a global graph.

Description

Technical Field

[0001] The present invention relates to the field of construction technology, and in particular, to a method for identifying construction drawings based on AI construction drawing recognition. Background Art

[0002] During the current process of identifying construction drawings, problems such as complex drawing information, non-standard annotations, and time-consuming and laborious manual recognition are often faced. Traditional drawing recognition methods often rely on manual interpretation, which is not only inefficient but also error-prone. These problems are particularly prominent when dealing with large-scale or complex construction projects. Therefore, there is an urgent need for a method that can efficiently and accurately identify construction drawing information to meet the development needs of the modern construction industry. The present invention is proposed based on this background, aiming to achieve automatic recognition of construction drawings through AI technology and improve recognition efficiency and accuracy. Summary of the Invention

[0003] The object of the present invention is to provide a method for identifying construction drawings based on AI construction drawing recognition.

[0004] To achieve the above object, the present invention adopts the following technical solutions: A method for identifying construction drawings based on AI construction drawing recognition, comprising the following steps: Train a variational autoencoder to map CAD markings to the latent space; Apply latent diffusion deformation to denoise the latent CAD code under the condition of the input image; Enhance the latent diffusion model by adding additional constraints and regularization; Introduce RLHF to implement the role of CAD code inspection. The code checker improves the effectiveness of denoising the latent code, uses the CAD compiler as an automatic checker, and classifies the code as valid or invalid; Fine-tune the diffusion model using the above classification set through direct preference optimization; Then extract the classification set and add identifiers, which are crucial for retrieving information from the source, and can generate evidence-based responses and start constructing a meta-graph in the later stage; Next, use the generative adversarial network GAN to perform adversarial training on the representation in the latent space to further improve the authenticity and diversity of the generated drawings.

[0005] Introduce a cross-domain transfer learning strategy, enabling the model to effectively identify and convert between construction drawings of different styles; By integrating the attention mechanism, the model can accurately capture the key details in the drawings.

[0006] Preferably, the autoencoder is trained to further extract drawing features. The autoencoder learns to compress the input drawing data into a low-dimensional latent representation through unsupervised learning and attempts to reconstruct the original input from this latent representation.

[0007] Preferably, during the recognition process, a processing mechanism for noisy latent CAD codes is adopted. Noisy latent CAD codes refer to the noise, interference, or irregular information that may exist in the drawing data.

[0008] Preferably, in the recognition process, we introduce the RLHF (Reinforcement Learning from Human Feedback) technique to implement CAD code checking. The RLHF technique conducts in-depth analysis and verification of CAD codes by simulating the decision-making process of human experts.

[0009] Preferably, after constructing the meta-graph, each classification set is scanned to develop a global graph that connects all the meta-graphs together. The nodes in these merged meta-graphs will be interconnected based on the link rules we used in the last paragraph.

[0010] Preferably, the details include structural connections, dimension markings, etc.

[0011] Preferably, the identifier is repeatedly extracted several times.

[0012] Preferably, by applying specific labels, the calculation is used to evaluate the similarity between two meta-graphs. The meta-graph with the highest similarity is selected for the merging operation. After the merging operation, the newly formed graph will inherit the structure and labels of the original meta-graph, ensuring the convenience of subsequent indexing. Then, new summary label information is assigned to the newly generated graph, and its similarity to other graphs is re-evaluated.

[0013] Preferably, during the process of fine-tuning the diffusion model, the effective and invalid CAD code samples in the classification set are used to fine-tune the diffusion model. Through this, the model can learn how to more effectively distinguish and process different types of CAD codes, thereby improving its accuracy and robustness in the building drawing recognition task.

[0014] The present invention has at least the following beneficial effects: The solution proposed in this design demonstrates a high degree of flexibility and scalability. First, by means of the method of iteratively merging meta-graphs, complex information in architectural drawings can be efficiently integrated to form a global graph, which facilitates subsequent drawing recognition. At the same time, the number of iterations is restricted to prevent excessive loss of detailed information, reflecting a delicate balance between the merging effect and efficiency. Second, by fine-tuning the diffusion model and training it with effective and ineffective CAD code samples, the recognition accuracy and robustness of the model for architectural drawings are significantly improved, enabling the model to better adapt to actual application scenarios. In addition, this solution also has good customizability, allowing users to adjust the number of iterations and fine-tuning parameters according to actual needs to achieve the best recognition effect. In summary, this solution has broad application prospects and important research value in the field of architectural drawing recognition. Description of the Drawings

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0016] Figure 1 It is a flowchart of the present invention. Detailed Embodiments

[0017] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0018] Refer to Figure 1 , a method for recognizing architectural drawings based on AI building drawing recognition, includes the following steps: S1. Train a variational autoencoder to map CAD markings to the latent space and further extract drawing features by training the autoencoder. The autoencoder learns, through unsupervised learning, to compress the input drawing data into a low-dimensional latent representation and attempts to reconstruct the original input from this latent representation. This process not only helps capture the core features of the drawings but also effectively removes redundant information, providing a more refined and effective feature representation for subsequent classification, recognition, and other tasks. By optimizing the reconstruction error of the autoencoder, its ability to extract drawing features can be continuously improved, laying a solid foundation for subsequent architectural drawing recognition; S2. Apply potential diffusion deformation to denoise the latent CAD code under the input image conditions. Through the processing mechanism of the noisy latent CAD code, the noisy latent CAD code refers to the noise, interference, or irregular information that may exist in the drawing data. If this information is not processed, it may affect the accurate recognition of the building drawing by the model. By designing specific preprocessing steps, the present invention can effectively identify and filter out these noisy latent CAD codes, ensuring that the drawing data input into the model is clean and accurate. This processing mechanism not only improves the stability of recognition but also further enhances the adaptability of the model in different complex drawing environments; S3. Enhance the latent diffusion model by adding additional constraints and regularization; S4. Introduce RLHF to achieve the role of CAD code inspection. The code checker is used to improve the effectiveness of denoising the latent code. The CAD compiler is used as an automatic checker to classify the code as valid or invalid. In the recognition process, we introduce RLHF, a reinforcement learning technique based on human feedback, to achieve CAD code inspection. The RLHF technique deeply analyzes and verifies the CAD code by simulating the decision-making process of human experts. It can identify potential inconsistencies, errors, or omissions in the drawing data, which are often difficult to discover through traditional rule matching or automated tools. Through the RLHF technique, we can ensure the accuracy and consistency of the CAD code, further improving the reliability and precision of building drawing recognition. This innovation not only enhances the model's processing ability for drawing data but also provides a strong guarantee for subsequent drawing parsing and recognition; S5. Use the above classification set to fine-tune the diffusion model through direct preference optimization to improve the generation quality and accuracy; S6. Then extract the classification set and add identifiers. The identifiers are crucial for retrieving information from the source and can generate evidence-based responses and start constructing a meta-graph in the later stage. After constructing the meta-graph, each classification set is scanned to develop a global graph that connects all the meta-graphs. The nodes in these merged meta-graphs will be interconnected based on the linking rules we used in the last paragraph. To this end, we calculate the distance between each pair of meta-graphs and sequentially merge the closest meta-graphs into larger entities. A summary is generated for each category, derived from the content of the meta-graph, thus forming a list of labels that concisely describe its main theme; S7. To improve the quality of extraction, reduce noise and variance, the identifiers are repeatedly extracted multiple times. This iterative method encourages the detection of any entities that might have been initially overlooked. After extracting the identifiers iteratively multiple times, we adopt an advanced filtering algorithm to further refine the extraction results, ensuring that only the most relevant and accurate identifiers are retained. This step is crucial for enhancing the performance and reliability of the overall recognition system. By reducing noise and variance, we can more precisely parse the complex information in architectural drawings, laying a solid foundation for subsequent building information modeling (BIM) construction and other applications. Additionally, this iterative extraction and filtering method also enhances the system's adaptability to different types and styles of architectural drawings, enabling it to be more widely applied in various practical scenarios.

[0019] S8. Next, a generative adversarial network (GAN) is used to perform adversarial training on the representations in the latent space to further enhance the authenticity and diversity of the generated drawings.

[0020] S9. A cross-domain transfer learning strategy is introduced to enable the model to effectively recognize and transform between architectural drawings of different styles; S10. By integrating an attention mechanism, the model can accurately capture key details in the drawings, including structural connections, dimension markings, etc., thereby improving the accuracy and integrity of recognition.

[0021] Among them, by applying specific tags, calculations are used to evaluate the similarity between two meta-graphs, and the meta-graph with the highest similarity is selected for the merging operation. After the merging operation, the newly formed graph will inherit the structure and tags of the original meta-graph, ensuring the convenience of subsequent indexing. Then, new summary tag information is assigned to the newly generated graph, and its similarity with other graphs is re-evaluated to explore the potential for further merging. This process can be executed iteratively until a single global graph is finally formed. However, as the summary tag information accumulates continuously, detailed information may gradually be lost, which reflects the need to balance between the merging effect and efficiency. To prevent excessive loss of detailed information, this study limits this iterative process to within 24 times.

[0022] During the process of fine-tuning the diffusion model, the diffusion model is fine-tuned using the valid and invalid CAD code samples in the classification set. Through this, the model can learn how to more effectively distinguish and process different types of CAD codes, thereby improving its accuracy and robustness in the architectural drawing recognition task. This step further enhances the model's ability to understand and process CAD codes in the latent space, laying a solid foundation for subsequent drawing recognition and applications. According to the above working process, it can be seen that this design has high flexibility and scalability. First of all, by means of the method of iteratively merging meta-graphs, complex information in architectural drawings can be efficiently integrated to form a global graph, which provides convenience for subsequent drawing recognition. At the same time, the number of iterations is limited to prevent excessive loss of detailed information, reflecting a delicate balance between the merging effect and efficiency. Secondly, by fine-tuning the diffusion model and training it with effective and invalid CAD code samples, the recognition accuracy and robustness of the model for architectural drawings are significantly improved, enabling the model to better adapt to the actual application scenarios. In addition, this design also has good customizability, and users can adjust the number of iterations and fine-tuning parameters according to actual needs to achieve the best recognition effect. In summary, this design has broad application prospects and important research value in the field of architectural drawing recognition.

[0023] The above has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and what is described in the above embodiments and the specification is only the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection required by the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for identifying construction drawings based on AI building drawing recognition, characterized in that, It includes the following steps: Train a variational autoencoder to map CAD markings to the latent space; Apply latent diffusion transformation to denoise the latent CAD code under the condition of the input image; Enhance the latent diffusion model by adding additional constraints and regularization; Introduce RLHF to implement the role of CAD code inspection. The code inspector improves the effectiveness of denoising the latent code. Use the CAD compiler as an automatic checker to classify the code as valid or invalid; Fine-tune the diffusion model using the above classification set through direct preference optimization; Then extract the classification set and add identifiers. The identifiers are crucial for retrieving information from the source, enabling the generation of evidence-based responses in the later stage and starting to construct the meta-graph; Next, use the generative adversarial network GAN to conduct adversarial training on the representations in the latent space to further enhance the authenticity and diversity of the generated drawings.

2. Introduce a cross-domain transfer learning strategy to enable the model to effectively identify and transform between architectural drawings of different styles; By integrating the attention mechanism, the model can accurately capture the key details in the drawings.

3. The method for identifying architectural drawings based on AI architectural drawing recognition according to claim 1, characterized in that, Train the autoencoder to further extract the drawing features. The autoencoder learns to compress the input drawing data into a low-dimensional latent representation through unsupervised learning and attempts to reconstruct the original input from this latent representation.

4. A method for identifying construction drawings based on AI construction drawing recognition according to claim 1, characterized in that, During the recognition process, through the processing mechanism of the noisy latent CAD code, the noisy latent CAD code refers to the noise, interference, or irregular information that may exist in the drawing data.

5. A method for identifying architectural drawings based on AI architectural drawing recognition according to claim 1, characterized in that, In the recognition process, we introduce RLHF, a human feedback-based reinforcement learning technique, to implement CAD code inspection. The RLHF technique deeply analyzes and verifies the CAD code by simulating the decision-making process of human experts.

6. A method for identifying architectural drawings based on AI architectural drawing recognition according to claim 1, characterized in that, After constructing the meta-graph, scan each classification set to develop a global graph that connects all the meta-graphs. The nodes in these merged meta-graphs will be interconnected based on the link rules we used in the last paragraph.

7. A method for identifying architectural drawings based on AI architectural drawing recognition according to claim 1, characterized in that, Details include structural connections and dimension markings.

8. A method for identifying architectural drawings based on AI architectural drawing recognition according to claim 1, characterized in that, The identifier is repeatedly extracted several times.

9. A method for identifying construction drawings based on AI building drawing recognition according to claim 1, characterized in that, By applying specific tags, the calculation is used to evaluate the similarity between two meta-graphs. Select the meta-graph with the highest similarity for the merging operation. After the merging operation, the newly formed graph will inherit the structure and tags of the original meta-graph, ensuring the convenience of subsequent indexing. Then, assign new summary tag information to the newly generated graph and re-evaluate its similarity with other graphs.

10. For an architectural drawing recognition method based on AI building drawing recognition as described in claim 1, during the process of fine-tuning the diffusion model, use the valid and invalid CAD code samples in the classification set to fine-tune the diffusion model. Through this, the model can learn how to more effectively distinguish and process different types of CAD codes, thereby improving its accuracy and robustness in the architectural drawing recognition task.

Images