An Intelligent Analysis Method of Drawings with Semantic Enhanced Topology Reconstruction

In the context of digital transformation within the engineering domain, intelligent parsing of drawings stands as a core technology for enabling automated information processing, confronted with critical challenges such as complex topological relationship parsing, multimodal semantic fusion, and unstructured semantic representation. This study proposes a method for intelligent parsing of large-scale engineering drawings based on semantically enhanced topological reconstruction. By leveraging Graph Neural Networks to construct topological relationship models of drawing elements, a cross-modal semantic alignment mechanism is utilized to integrate multimodal information encompassing images, texts, and symbols. Furthermore, a vectorized representation model is designed to capture deep engineering semantic features. Experimental results demonstrate a $23.7 \%$ improvement in accuracy for complex topological relationship recognition compared to traditional methods. This significantly enhances the efficiency and accuracy of drawing parsing, offering an effective solution for the intelligentization of engineering design and construction.