JUHCCR-v1: a database for hand-drawn electrical and electronics circuit component recognition.

Automatic electrical and electronics component/symbol recognition from hand-drawn circuits is a challenging research problem. However, the literature survey reveals that there has been no significant progress in this domain. One possible reason for this might be a lack of publicly available datasets. To this end, in this work, we have developed a dataset, called JUHCCR-v1, which comprises 20 different hand-drawn circuit components that are commonly found in electrical and electronic circuits. Additionally, we have prepared a synthetic dataset having different variations (like orientations, stroke lengths, and distortions) of collected circuit components that may occur while extracting the components from an entire hand-drawn circuit diagram. This augmented dataset with the original ones would help train the deep learning based circuit component recognition algorithms. In order to provide a base result on this dataset, we have designed a weighted ensemble-based hand-drawn circuit component recognition method applied to snapshots of the convolutional block attention module-aided DenseNet-121 architecture. This benchmarking method achieves an accuracy of 91.15% on test set images. All datasets prepared here, along with codes, are made publicly available for the research community at: https://github.com/AyushRoy2001/Circuit-Component-Analysis .