Vision-Based Topology-Consistent Structural Parsing of Hand-Drawn Circuit Diagrams.

Area of Science:

• Computer Vision
• Electrical Engineering
• Image Processing

Background:

• Hand-drawn circuit diagrams are widely used but difficult to process digitally.
• Challenges include irregular strokes, discontinuities, and imaging artifacts.
• Existing methods often struggle with accurate topology recovery.

Purpose of the Study:

• To develop a robust method for recovering circuit topology from images of hand-drawn diagrams.
• To enable conversion of visual circuit representations into machine-readable netlists.
• To facilitate downstream simulation and verification of hand-drawn circuits.

Main Methods:

• A topology-consistent structural parsing framework was proposed.
• Integrated multi-source visual perception and connectivity reasoning.
• Employed explicit endpoint semantic recovery for component analysis.

Main Results:

• Achieved a 95.14% strict image-level end-to-end success rate on a benchmark of 1317 diagrams.
• Successfully exported recovered structures as Simulation Program with Integrated Circuit Emphasis (SPICE)-compatible netlists.
• Demonstrated practical vision-based workflow for circuit image processing.

Conclusions:

• The proposed framework effectively recovers circuit topology and semantics from hand-drawn images.
• Enables automated conversion of visual circuit data for simulation and verification.
• Supports a practical workflow for digitizing hand-drawn circuit diagrams.