Next Article in Journal
Hierarchical Vision–Language Pre-Training with Freezing Strategy for Multi-Level Semantic Alignment
Previous Article in Journal
Human-Centric Microgrid Optimization: A Two-Time-Scale Framework Integrating Consumer Behavior
 
 
Article
Peer-Review Record

Research on Intelligent Verification of Equipment Information in Engineering Drawings Based on Deep Learning

Electronics 2025, 14(4), 814; https://doi.org/10.3390/electronics14040814
by Zicheng Zhang 1,* and Yurou He 2
Reviewer 1:
Reviewer 2:
Reviewer 3: Anonymous
Electronics 2025, 14(4), 814; https://doi.org/10.3390/electronics14040814
Submission received: 19 January 2025 / Revised: 14 February 2025 / Accepted: 17 February 2025 / Published: 19 February 2025
(This article belongs to the Section Artificial Intelligence)

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper addresses the task of automatic table structure recognition and understanding in engineering drawings, which is of utmost importance during the summarization and display of information in the digitalization process. In this paper, an intelligent verification approach is proposed by integrating table object recognition with YOLOv10, an Improved LORE algorithm for table structure extraction, RapidOCR for text recognition, and semantic similarity matching using BERT. 

 

The manuscript limits the LORE algorithm to handling only small images. While it improves on this, it does not indicate the maximum size of images that can be handled.

 

Since the solution integrated several complicated algorithms- YOLOv10, LORE, RapidOCR, and BERT- the learning curve would likely be steep and complex to implement, thus a potential obstacle in practical scenarios.

 

 

While the paper introduces advanced semantic similarity calculations using BERT and the improved cuckoo search algorithm, handling a variety of semantic variations in equipment names across different documents is not deeply explored in real-world efficacy.

 

No concrete error analysis was conducted or considered for the different kinds of detected errors, which would likely shed more light on which future improvements or fine-tunings are actually necessary in this system.

 

Although some comparisons are made with state-of-the-art methods like pp structure v2 and original LORE, the manuscript would benefit from a more extensive comparison to more recent or competing technologies that allow highlighting the advantages and shortcomings of the proposed method.

 

 The manuscript lacks detailed information on the size, diversity, and source of the datasets used for training and validation. This information is important in understanding the robustness of the proposed approach.

 

The study has not discussed the proposed system's scalability or adaptability to other data types or sectors other than engineering drawings, which might limit its broader applicability.

Author Response

Thank you to the reviewers for their hard work. We have made point-by-point revisions, and the details can be found in the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The novel contributions of this approach should be better addressed and the empirical results should better consider the eventual risk of bias. The research gap is not properly identified and this paper does not offer a Discussion section.

Main issues:

1. The paper stresses the existence of limitations in table processing but it is not clear the research gap this study addresses considering previous research in the same field.

2. Introduction section should present the structure of the paper.

3. Authors note the use of deep learning technology but it is not provided a global vision about the potentialities offered by this technology. Authors only give some isolated examples, which is not enough.

4. Why deep learning in this context is more relevant than machine learning approaches?

5. Authors claim the use of an intelligent verification approach for table structure recognition in engineering drawings. What are the intelligent components?

6. why this model is only applied to engineering drawings?

7. It would be important to emphasize the novel contributions of this study.

8. This model only works with 640x640 pixels size?

9. Where did you get the training dataset?

10. How did you categorized the target categories in Table 1? Are there uniformly represented?

11. It would be crucial to offer a Discussion section.

12. Practical contributions of this study are not explored.

13. Authors should also discuss the limitation of their approach.

14. Future research directions are very superficial and should be revised to give more relevant insights.

15. Number of references should be increased.

Author Response

Thank you to the reviewers for their hard work. We have made point-by-point revisions, and the details can be found in the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

1) The performance of mAP50-95 is significantly lower compared to precision, recall, and mAP50. Please explain the reasons behind this discrepancy.

2) In addition to the performance metrics currently presented, please provide additional evaluation metrics to offer a more comprehensive comparison.

3) While YOLOv10 and BERT-based approaches demonstrate high accuracy, there is no discussion regarding real-time performance in processing engineering drawings. Optimization of computation speed is a critical aspect that must be addressed; please include this in your discussion.

4) The authors' proposed method lacks detailed explanations of its practical impact on the engineering field. Please elaborate on how this method could influence practical applications in the industry.

5) The limitations of this study and suggestions for future research have not been adequately discussed. Please provide detailed content to address these aspects.

Author Response

Thank you to the reviewers for their hard work. We have made point-by-point revisions, and the details can be found in the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have revised the manuscript. The manuscript may be accepted.

Author Response

Thanks for the reviewer's recognition

Reviewer 2 Report

Comments and Suggestions for Authors

I recognize important improvements in the manuscript. However, there are still critical issues that were not properly addressed:

  1. Introduction section only uses one reference. Motivation and research gap is not properly explored and justified from a scientific perspective.
  2. Discussion section is relevant considering a technical perspective. However, from a scientific perspective it is a weak section. It is important to highlight the main contributions of this work considering previous published works. Please use references to support this discussion.
  3. Conclusion section can be extended and better organized. Authors can use sub-sections to better organize their ideas.
  4. Despite the improvements in the number of references, there is space and possibilities to increase them.

Author Response

Thanks for the reviewer's comments, we have revised the article again, please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript has improved significantly.

Author Response

Thanks for the reviewer's recognition

Round 3

Reviewer 2 Report

Comments and Suggestions for Authors

the paper can be accepted at this phase, after the improvements made by the authors. 

Back to TopTop