A review of deep learning methods for digitisation of complex documents and engineering diagrams.

Jamieson, Laura; Moreno-Garcia, Carlos Francisco; Elyan, Eyad

Abstract

This paper presents a review of deep learning on engineering drawings and diagrams. These are typically complex diagrams, that contain a large number of different shapes, such as text annotations, symbols, and connectivity information (largely lines). Digitising these diagrams essentially means the automatic recognition of all these shapes. Initial digitisation methods were based on traditional approaches, which proved to be challenging as these methods rely heavily on hand-crafted features and heuristics. In the past five years, however, there has been a significant increase in the number of deep learning-based methods proposed for engineering diagram digitalisation. We present a comprehensive and critical evaluation of existing literature that has used deep learning-based methods to automatically process and analyse engineering drawings. Key aspects of the digitisation process such as symbol recognition, text extraction, and connectivity information detection, are presented and thoroughly discussed. The review is presented in the context of a wide range of applications across different industry sectors, such as Oil and Gas, Architectural, Mechanical sectors, amongst others. The paper also outlines several key challenges, namely the lack of datasets, data annotation, evaluation and class imbalance. Finally, the latest development in digitalising engineering drawings are summarised, conclusions are drawn, and future interesting research directions to accelerate research and development in this area are outlined.

Citation

JAMIESON, L., MORENO-GARCIA, C.F. and ELYAN, E. 2024. A review of deep learning methods for digitisation of complex documents and engineering diagrams. Artificial intelligence review [online], 57(6), article number 136. Available from: https://doi.org/10.1007/s10462-024-10779-2

• Journal Article Type: Article
• Acceptance Date: Apr 24, 2024
• Online Publication Date: May 9, 2024
• Publication Date: Jun 30, 2024
• Deposit Date: May 12, 2024
• Publicly Available Date: May 23, 2024
• Journal: Artificial intelligence review
• Print ISSN: 0269-2821
• Electronic ISSN: 1573-7462
• Publisher: Springer
• Peer Reviewed: Peer Reviewed
• Volume: 57
• Issue: 6
• Article Number: 136
• DOI: https://doi.org/10.1007/s10462-024-10779-2
• Keywords: Deep learning; Object detection; Engineering diagram; Piping and instrumentation diagram; Convolutional neural networks
• Public URL: https://rgu-repository.worktribe.com/output/2332495