Enhanced Recognition of Hand-Drawn Circuit Components Using Fast R-CNN and Faster R-CNN

There has been a growing interest in digitizing hand drawn circuits. The first step in this regard would be to identify the different components drawn in the circuit. While many approaches exist in literature, they have been performed for cases where the number of components is minimal. In this paper, we propose two approaches for identifying hand drawn electronic circuit components. The first approach makes use of Fast Region Based Convolutional Neural Network (Fast R-CNN) for identification where as the second approach utilises Faster Region Based Convolutional Neural Network (Faster R-CNN). For experimentation purpose, we have created a dataset of 22 electronic components such as resistors, inductors, transistors and some basic gates which are hand drawn and comprising of 4235 images in total. The experiments show that the Faster R-CNN outperforms Fast R-CNN by $\mathbf{1. 5 \%}$.