An Automatic Recognition Method of Electrical Wiring Diagrams Based on Optimized YOLO11 and CAD Automatic Recognition Technology

Diagram-model consistency of electrical wiring diagrams remains a core challenge for multi-source data collaboration in smart substations, with structural discrepancies between SVG diagrams (design institutes) and CIM/E models (dispatch systems) causing 22% manual comparison errors. To address limitations of existing methods in small-target detection (e.g., 0.08%-area isolating switches), imbalanced component distribution (<10 samples for rare components), and joint topology-text parsing, this work proposes an intelligent framework integrating enhanced YOLO11 with CAD parsing. Key contributions include: an adaptive sliding window mechanism for small-target recognition, hierarchical feature fusion with CBAM attention to distinguish geometrically similar elements, and a DXF spatial rule engine for connection extraction. Evaluations on real-world grid datasets achieve 93.6% mAP@50 (21.1-point improvement over baseline YOLO11), 86% rare-component recognition, 100% text extraction accuracy, 99.4% connection accuracy, and 5.2s/image processing (3.04× faster than conventional approaches). This framework enables comprehensive collaborative parsing of 28 component categories with topological relationships, providing effective technical support for power system digitalization.