VectorGraphNET: Graph Attention Networks for Accurate Segmentation of Complex Technical Drawings

This study introduces a new approach to extracting vector data from technical drawings in Portable Document Format (PDF) format and analyzing them semantically by employing graph attention networks. The proposed method involves converting PDF files into Scalable Vector Graphics (SVG) format and creating a feature-rich graph representation, which captures the relationships between vector entities using geometrical information. We then apply a graph attention transformer with hierarchical label definition to achieve accurate line-level semantic segmentation. Our approach is evaluated on two data sets, including the public FloorplanCAD data set, which achieves state-of-the-art results on weighted F1-score (89%), surpassing existing methods. Our vector-based method offers a scalable large-scale technical drawing analysis solution, requiring significantly less graphical processing unit (GPU) power than does current state-of-the-art techniques. Our method performs better semantic segmentation tasks, effectively extracting meaningful information from technical drawings. This enables reliable line-level classification, even for complex drawings such as architectural floorplans with overlapping structural and annotation layers, and opens up new applications, such as automated drawing processing. Additionally, our method can improve existing workflows in the Architecture, Engineering, and Construction (AEC) industry, including automated building information modeling (BIM) and construction planning.