Vectorization for Tower Sketches Using Large Model and Deep Learning


== Abstract ==

<p>Background:3D modeling of transmission towers from hand-drawn sketches is occasionally needed in the exploration stage, construction stage, or operation of transmission lines. To achieve automatic 3D modeling, extracting standard vectors from the sketches is very important, however, this is still lacking for real applications. Methods: In this article, we propose a method for automatically converting the hand-drawn design drawings of a tower into standard vector diagrams. The network ControlNet is introduced to control the image generation model Stable Diffusion, to convert hand-drawn sketches into standard line-structural diagrams. Then, a vectorization network is used to vectorize the line structure, and some designed optimization modules are proposed to refine the vector extraction results. Results: Experiments prove that our method can output the most accurate and robust vectorization results compared with some state-of-the-art methods. Conclusion: The proposed method is automatic and does not need manual operations, which can bring convenience for the intelligent design and modeling of transmission towers. Furthermore, it can also provide technical support for developing relevant digital products for the smart grid.OPEN ACCESS Received: 31/10/2024 Accepted: 13/02/2025 Published: 30/05/2025</p>



== Document ==
<pdf>Media:Draft_Sanchez Pinedo_329555871-4593-document.pdf</pdf>