Binary Neighborhood Coordinate Descriptor for Circuit Board Defect Detection

Round 1

Reviewer 1 Report

The authors' proposal involves a novel method for detecting circuit board defects in an industrial environment. The quality of the presentation is good, involving tables and figures which improves the overall merit. The idea seems interesting, however, some weaknesses are evident and should be solved.

 

First, the bibliography seems not to correctly cover the overview of the treated topic. Most of the references are very old. We suggest the authors to improve it by adding some more recent works about industrial products defects detectors, e.g.;

Gao, Hongbin, et al. "A deep convolutional generative adversarial networks-based method for defect detection in small sample industrial parts images." Applied Sciences 12.13 (2022): 6569.

Avola, Danilo, et al. "Real-time deep learning method for automated detection and localization of structural defects in manufactured products." Computers & Industrial Engineering 172 (2022): 108512.

Deng, Honggui, et al. "Industrial laser welding defect detection and image defect recognition based on deep learning model developed." Symmetry 13.9 (2021): 1731.

Related to this problem, we also suggest the authors to uniform the bibliography style.

 

Probably related to the previous weakness, the idea seems to involve only image-processing techniques, naming extremely well-known approaches, such as SURF, ORB, RANSAC, and many others. It could be possible that the exploitation of those techniques only (without machine or deep learning) could be not enough competitive in 2023. We suggest the authors to strongly motivate their choices in order to highlight why their method could be considered a novelty even if the proposed pipeline exploits only outdated algorithms.

 

There is no mention of an involved dataset in the study. The only information about the boards used for tests is at line 438 without any further information. A self-comparison with different algorithms could provide a significant scientific result at least if a public dataset is involved. Otherwise, a comparison with other similar systems could be introduced. We suggest the authors to improve the overall scientific soundness of the work with one of those two suggestions.

 

If the authors solve those issues, the manuscript could be eligible for publication.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

This manuscript describes a Binary Neighborhood Coordinate Descriptor for circuit board defects detection. The manuscript requires revisions.

1) Per line 145-150, line 172-180, figure 8, and line 325-336, the authors use plain language to describe the components of the system. However, for scientific writing purpose, the authors are required to revise them to algorithms (in pseudo code) to include necessary mathematical forms/derivations. 

2) As for the section 2.3, the design of BNCD seems to heuristic. The authors are required to add motivations (from physical perspectives) regarding the design of BNCD. Why this design is unique/useful and why its performance is better than the state-of-the-art? 

3) The authors should be aware when the BNCD fails and what reasons might cause such failure. A failure analysis are required in this discussion. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors satisfied all the requests and introduced the missing parts of the first submission. This version of the manuscript has been corrected according to the needs.

Author Response

Dear reviewer,
  Thank you very much for your comments and professional advice. The article has been upgraded after being revised in accordance with your suggestions. We appreciated for your warm work earnestly. Once again, thank you very much for your time and attention.

Best wishes.
Jiaming Zhang

Reviewer 2 Report

The reviewer thanks the authors to make such changes. The manuscript is getting better, but there are several things the authors have to further revise. 

Regarding my previous comment 3): 

The added two lines discussion are not enough. The authors are required to provide a separate section to analyze why BNCD fails. This section will cover some specific experimental examples with exact matching results about such failures, such as Figure 11 but showing wrong matching results, etc. 

Regarding my other comments:

Please make sure the index line in respond letter matches the line in manuscript. For example, as the authors responded, "We explained why BNCD is useful from a physical perspective in revision mode lines 335-342", but it seems the line 335-342 in about the RANSAC. Please update the line index in respond letter accordingly.

Also, as for my previous comment 2): "As for the section 2.3, the design of BNCD seems to heuristic. The authors are required to add motivations (from physical perspectives) regarding the design of BNCD. Why this design is unique/useful and why its performance is better than the state-of-the-art?"

The question here is the authors should provide general motivations regarding the design of BNCD. If I rephase the language, the question is how did the auhors come up with such design and why did the authors think this design is useful? Please clarify. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 3

Reviewer 2 Report

The reviewer thanks the authors to make such changes. The manuscript looks good and it can be accepted in the current form.