Modern Deep Learning Approaches for Symbol Detection in Complex Engineering Drawings

• Published in: 2022 International Conference on Digital Transformation and Intelligence (ICDI) pp. 121 - 126
• Main Authors: Bhanbhro, Hina, Hooi, Yew Kwang, Hassan, Zaira, Sohu, Najamuddin
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.12.2022
• Subjects: Deep learning; Engineering drawings; Industries; Instruments; Learning systems; object detection; symbol detection; Symbols; Training
• ISBN: 9798350396980
• DOI: 10.1109/ICDI57181.2022.10007281

Due to the rapid advances in object detection methods, object recognition in the engineering drawings has gained researchers' attention in recent years. Usually, engineering drawings are analysed using traditional object detection methods. However, complex structures and their connections sometimes make it challenging for the traditional object detection architectures to process the engineering drawings. With the rapid development in deep learning models, more powerful methods- which are able to interpret high-level and complex features- are introduced to address the challenges exhibited by the traditional methods. These models have different training strategies and thus portray varied behaviors. In this paper, we present a comparative analysis of the deep learning models for object recognition in engineering drawings. Furthermore, real-life scenario on how to process and analyses information from specific engineering drawings, namely, piping and instrumentation diagrams (P&IDs) is discussed in details. Finally, future directions are presented which can serve as research guidelines to work in this domain.