Submit to this Journal Review for this Journal Propose a Special Issue

Article Menu

Printed Edition

A printed edition of this Special Issue is available at MDPI Books....

Share Help Cite Discuss in SciProfiles

Open AccessArticle

Peer-Review Record

A Self-Supervised Model Based on CutPaste-Mix for Ductile Cast Iron Pipe Surface Defect Classification

Sensors 2023, 23(19), 8243; https://doi.org/10.3390/s23198243

by Hanxin Zhang, Qian Sun and Ke Xu^*

Reviewer 1: Anonymous

Reviewer 2: Anonymous

Reviewer 3: Anonymous

Sensors 2023, 23(19), 8243; https://doi.org/10.3390/s23198243

Submission received: 24 August 2023 / Revised: 27 September 2023 / Accepted: 29 September 2023 / Published: 4 October 2023

(This article belongs to the Special Issue Sensors Data Processing Using Machine Learning)

Round 1

Reviewer 1 Report

Surface defect monitoring is quite important to ensure the safe operation of pipeline. This paper proposes a CutPaste-based approach for DCIP surface defect monitoring. The research is interesting, however, the following comment should be deal with before acceptance.

i. Please check the writing, as I can see some gramma mistakes.

ii. CutPaste is not a novel approach, and there are also papers try to use this approach for pipeline surface defect monitoring. So what is the contributions of the paper compared with existing approach? It should be more clear.

iii. What are the advantages of this approach cpmpared with other online surface inspection approach?

iv. Please compare the approach with advanced machine vision approach.

Author Response

Dear reviewers:

Thanks very much for your careful review and positive comments. We have substantially revised our manuscript sensors-2601049 entitled 'A Self-Supervised Model based on CutPaste for Ductile Cast Iron Pipes Surface Defect Classification', after reading the constructive suggestions provided by the three reviewers.

Please note that we have replaced 'CutPaste' in the title with a more accurate term, 'CutPaste-Mix.' So, the new title is 'A Self-Supervised Model based on CutPaste-Mix for Ductile Cast Iron Pipes Surface Defect Classification'. We believe this change is necessary to express our approach more clearly to readers.

In addition to the suggestions provided by the reviewers, we have also made some necessary revisions to the formulas and text to better clarify our research content. We have highlighted the revisions for your clear review. Certification for the clear and separate indications of our modifications is attached as the following. Please feel free to contact us with any questions and we are looking forward to your consideration.

Comment1: Please check the writing, as I can see some gramma mistakes.

ANSWER: Thank you for point that out, we have carefully checked and corrected grammatical mistakes.

Comment2: CutPaste is not a novel approach, and there are also papers try to use this approach for pipeline surface defect monitoring. So, what is the contributions of the paper compared with existing approach? It should be clearer.

ANSWER: We agree that CutPaste is not a novel method. However, our research goal is to apply the CutPaste-Mix (an improved CutPaste strategy) to the task of anomaly classification, rather than directly conducting defect detection. We have reviewed relevant literature and have not found a method or scenario that is entirely identical to ours.

Therefore, we have added relevant explanations and justifications in the article. For details, please refer to highlighted text in the second and fourth paragraphs of Introduction. In section 2.2, we provide a clearer and more detailed explanation of the specifics of our method.

Comment3: What are the advantages of this approach compared with other online surface inspection approach?

ANSWER: We built a binary classification model in order to better collect the defect data, the model can be trained using only normal surface data. It’s important to note that our research focus is on anomaly classification, which constitutes a distinct research domain from defect detection. Therefore, we believe that it's not necessary to conflate them. We believe that some expressions may have caused misunderstanding, and deeply apologize for that.

Comment4: Please compare the approach with advanced machine vision approach.

ANSWER: In the experiment, we have already compared CutPaste-Mix with highly regarded binary classification models: Deep One-Class (DOCC), Uniformed-Student, and Patch-SVDD.

Reviewer 2 Report

The authors address the problem of online surface inspection system particularly the fact that the assessment of gathered images would demand a significant amount of time. They propose a self-supervised binary classification algorithm to recognize images of ductile cast iron pipes (DCIP) showing damages based on machine vision and deep learning. The authors construct their own anomaly classification models, and achieve good results.

The paper is well written and structured. The authors employed modern techniques. Hence, the research work and the achieved results have some good merits for publication. There are only few, mostly technical aspects to be addressed:

1) All used abbreviations should be introduced when used for the first time and this is valid even if the authors believe that some abbreviations are well-known (for instance, AUC).

2) When using references, do not put them next to the words, but an empty space between them. Hence, instead of "...quality[1].", write "...quality [1]."

3) Be consistent when writing words. In the beginning, you wrote CutPaste, later Cutpaste.

4) When referring to figures, do not put a dot after the figure number. Hence, instead of "Figure 2.", write "Figure 2". ("Figure 2. showcases...")

5) The lines after equations starting with "where" should not be indented (this is not a new paragraph) and should actually start with a small letter (as it is continuation of the same sentence). Alternatively, if it is a new sentence, then start with "Here, ..."

6) Figure 9 is on page 13, but the authors refer to it on page 14 for the first time. A figure should appear only after it was referred to in the text for the first time.

7) Again, consistency in writing is important. Line 382: "...the results reached a remarkable 99.4 AUC." Number 99.4 is written partly italic, partly by using normal font. Only normal font should be used, as it is the case with all other numbers.

8) Conclusions should better emphasize the limitations of the proposed technique. They are partly mentioned as an incapability of the model to entirely distinguish between noise interferences and actual defects, but this should be given a bit more attention.

Author Response

Dear reviewers:

Comment1: All used abbreviations should be introduced when used for the first time and this is valid even if the authors believe that some abbreviations are well-known (for instance, AUC).