PSNet: Parallel-Convolution-Based U-Net for Crack Detection with Self-Gated Attention Block
Round 1
Reviewer 1 Report
The work in general terms is interesting and presents an improvement to the results found in the most frequently available literature.
It is requested:
Define all acronyms appropriately, e.g. UNet .
When reference [2] Marques is used, the number is incorrect, it should be number [3].
In line 48 when the YOLO v3 method is mentioned, it is convenient to incorporate a bibliographic reference.
Please check that all acronyms and abbreviations are defined before being used in the document, in order to facilitate the reading of a wider audience, such as the journal in question.
In the figure caption 1 change the final comma to a period.
In line 115 they do not define what EISeg is.
A period is missing in the mathematical expressions (1) and (2), and include commas at the end of expressions (3) and (4).
Revise the verb conjugation at the end of line 219.
In the first paragraph of section 2.5 they abuse the use of capital letters when naming the transformations.
To clarify in line 232 the intervals in which are measured angles are in sexagesimal degrees.
A full stop is missing in section 2.6.
In line 249 please specify with a reference where Google proposes the activation function.
A period is missing in the expression (5).
Please describe the acronyms and abbreviations on line 272 and following.
Idem with line 284.
It is missing to describe which variable is being presented in tables 1,2,3 and 4. It can be inferred which one but it should be explicitly stated.
Line 310 please include a reference.
In table 3, with MSE the author want to refer to the mean square error, it must be inferred, it is not defined.
There is a typo in line 303.
Please check that all bibliographic references are complete, for example reference number 13, line 394 is incomplete.
A light language revision should be done.
Author Response
Point 1： Define all acronyms appropriately, e.g. UNet ..
Response 1：I have changed all UNet into U-Net
Point 2: When reference [2] Marques is used, the number is incorrect, it should be number [3].
Response 2: I have revised it in my paper
Point 3: In line 48 when the YOLO v3 method is mentioned, it is convenient to incorporate a bibliographic reference.
Response 3: I have add a reference
- Nie M , Wang C . Pavement Crack Detection based on yolo v3[C]// 2019 2nd International Conference on Safety Produce Informatization (IICSPI). 2019.
Point 4: Please check that all acronyms and abbreviations are defined before being used in the document, in order to facilitate the reading of a wider audience, such as the journal in question..
Response 4: I have revised it in my paper
Point 5: In the figure caption 1 change the final comma to a period.
Response 5: I have revised it in my paper
Point 6: In line 115 they do not define what EISeg is.
Response 6: I have revised it in my paper, EISeg (an image segmentation labeling tool provided by Baidu)
Point 7: A period is missing in the mathematical expressions (1) and (2), and include commas at the end of expressions (3) and (4).
Response 7: I don't quite understand what this means
Point 8: Revise the verb conjugation at the end of line 219..
Response 8: I have revised it in my paper
Point 9: In the first paragraph of section 2.5 they abuse the use of capital letters when naming the transformations.
Response 9: I have revised it in my paper
Point 10: To clarify in line 232 the intervals in which are measured angles are in sexagesimal degrees.
Response 10: I have revised it in my paper
Angle Rotation: Traditionally, angle rotation is used to augment training datasets through rotating images into different angles. Usually, since crack images are invariant to random rotation, thus we rotate all images at degree angle of 10, 30, 60, 90, 110, 140, 170 respectively to generate new images.
Point 11: In line 249 please specify with a reference where Google proposes the activation function.A period is missing in the expression (5).
Response 11: I have revised it in my paper
Point 12: Please describe the acronyms and abbreviations on line 272 and following.Idem with line 284.
Response 12: I have revised it in my paper
FCN: a fully convolutional neural network
ConvNet : a deep convolutional neural network
Split-Attention Network: a channel-wise attention based network
Cascaded Attention DenseUNet: an attention based network with global attention and core attention
ECA-Net: a lightweighted channel attention based convolutional neural network
DWTA-UNet: a U-Net based network with discrete wavelet transformed image features.
Point 13: It is missing to describe which variable is being presented in tables 1,2,3 and 4. It can be inferred which one but it should be explicitly stated.
Response 13: I don't quite understand what this means
Point 14: Line 310 please include a reference..
Response 14: I have revised it in my paper
Point 15: In table 3, with MSE the author want to refer to the mean square error, it must be inferred, it is not defined.
Response 15: I have revised it in my paper
Point 16: There is a typo in line 303.
Response 16: I have revised it in my paper
Point 17: Please check that all bibliographic references are complete, for example reference number 13, line 394 is incomplete.
Response 16: I have revised it in my paper
Author Response File: 
Author Response.pdf
Reviewer 2 Report
A very good paper, please refer to the attached file for reflection. All the best.
Comments for author File: Comments.pdf
Author Response
Dear reviewer, thank you for pointing out the issue. I have corrected all the issues you pointed out in the article, and highlighted all modifications in the article,
Please see the attachment
Author Response File: 
Author Response.docx
Reviewer 3 Report
In this paper, the authors presented a crack detection model based on a proposed PSNet model. In this model, a Parallel Convolution Module (PCM) is designed to avoid feature information loss caused by down-sampling. This means that more low-level features could be used for the classification. Overall, the methodology and experiments are well conducted, and the results are fascinating. However:
- The manuscript requires serious revision and proofreading because it contains many errors in writing and organization.
- The paper’s organization should be added in the last part of the introduction.
- When putting an equation, all parameters should be defined. Check all equations.
- More details should be added about dataset splitting and protocols of evaluation.
- The references are incompletes and should be written according to the MDPI format and professionally.
- Authors should add more perspectives.
The manuscript requires serious revision and proofreading because it contains many errors in writing and organization.
Author Response
Dear reviewer, thank you for pointing out the issue. I have corrected all the issues you pointed out in the article, and highlighted all modifications in the article,
Please see the attachment
Author Response File: Author Response.docx
Reviewer 4 Report
The Manuscript is very well prepared and within the scope of the journal.
There are just a few points which require attention.
- please present more detailed analyses of the dataset used, and point out the differences.
- please explain what is the contribution of this stud to the field while there are many works done in this field.
- please explain what are the problems with this approach. I mean in what type of figures this algorithm is wrong, what type of cracks are missdetected or algorithm stated the crack when it was not.
-please represent the contribution to the field of knowledge in the conclusion section
i do not feel qualified to do this.
Author Response
Please see the attachment
Author Response File: Author Response.docx
Reviewer 5 Report
This article shows the possibility of using modules PCM and SGAB in Unet convolutional neural network for Crack Detection.
The paper provides an interesting and valuable contribution.
The article is well written, good designed and the problem that the authors are investigating is relevant.
In most, the structure of the article meets the classical requirements.
The article title is clear.
Keywords are chosen correctly and match the direction of the research.
The abstract contains mandatory attributes, namely the prerequisites, a description of the methods used, a presentation of the results and their comparison with similar studies, and the main conclusions.
The authors have clearly described the introduction and methodology, and I didn't have any questions to these sections.
However, there are points that could be improved.
The authors used an interesting approach to the presentation of the results of their research, namely a comparison with the results of other articles.
The main purpose of creating such models is their further practical use for the identification of cracks.
So, I think that in the "Results" section, the authors need to show an example of how the trained model classifies real "crack"/"non-crack" images.
Author Response
Please see the attachment
Author Response File: Author Response.docx
Round 2
Reviewer 3 Report
The authors should seriously consider my previous comments. Otherwise, I will reject the manuscript.
The manuscript requires a proofreading.
Author Response
Please see the attachment
Author Response File: Author Response.docx
Round 3
Reviewer 3 Report
The manuscript can be accepted in the current form.
No comment.
