CADRNet: geometric figure recognition in architectural drawings via dynamic feature pyramid and cross-architecture adapter

As the automatic recognition of symbols in computer-aided-design (CAD) drawings is a key technical bottleneck for realizing the digital transformation of engineering design, the main challenges faced by current research are the multi-scale variability of engineering symbols and their complex contextual dependencies. In this study, CADRNet, a framework for engineering drawing symbol recognition based on a CNN-Transformer hybrid architecture, is proposed. The model solves the above problems through two core innovative mechanisms: (1) Dynamic Feature Pyramid Network (DFPN), which uses the attention mechanism to guide the coordinated extraction and integration of multi-scale features to effectively cope with the scale variation of symbols; (2) Bidirectional Cross-Architecture Adapter (CTAdapter), which realizes the bidirectional interaction and deep fusion between the local texture features extracted by CNN and the global context information captured by Transformer. Experiments on the standard CAD symbol recognition benchmark show that CADRNet significantly outperforms existing methods in three key evaluation indicators, with PQ, SQ and RQ of the test set reaching 0.679, 0.876 and 0.678 respectively, which are 10.2%, 10.1% and 5.8% higher than the state-of-the-art CADTransformer model. These results verify the effectiveness and advancement of the proposed method in the field of automated analysis of engineering drawings.