Hand-drawn Electric Circuit Diagrams Recognition using Deep Learning

This paper tackles component identification in hand-drawn electrical circuit diagrams by employing Region-based Convolutional Neural Networks (R-CNN). This paper introduces a new offline circuit recognition system for hand-drawn diagrams, including a new annotated dataset. We surveyed 200 engineers who possess knowledge of electric diagrams with a three-page document to propose an annotated dataset containing 6273 symbols distributed for 12 classes and 728 diagrams distributed for three classes. Our proposed Faster RCNN model recognizes components with an accuracy of $75.9 \%$ mAP.