Pixel-Level Concrete Crack Segmentation Using Pyramidal Residual Network with Omni-Dimensional Dynamic Convolution

Round 1

Reviewer 1 Report

see appendix

Comments for author File: Comments.pdf

Author Response

Dear Editor and Reviewers: On behalf of my co-authors, we are very grateful to you for giving us an opportunity to revise our manuscript. we appreciate you very much for your positive and constructive comments and suggestions on our manuscript entitled “Pixel-Level Concrete Crack Segmentation Using Pyramidal Residual Network with Omni-Dimensional Dynamic Convolution” (ID: processes-2175989). We have studied reviewers’ comments carefully and tried our best to revise our manuscript according to the comments. The following are the responses and revisions I have made in response to the reviewers' questions and suggestions on an item-by-item basis. Thanks again to the hard work of the editor and reviewer! Comment No. 1: Author should give the training time or a comparison of network training time. Response: Thanks to Reviewer for reminder, we add and compare the training times in Section 4.6. Comment No. 2: In the section of 4.3. Network Results, the number of training samples, verification sample and test sample, Response: Thanks to Reviewer for reminder, we add a specific description of dataset partitioning in Section 4.1. Comment No. 3: The language has some problems. Authors should check the language and improve the language quality. Response: Thanks to Reviewer for reminder, we were accompanied by a professional who carefully corrected the language issues. Comment No. 4: Authors should state the scope of application of the method, i.e. what is the smallest crack that can be segmented. Response: Thanks to Reviewer for reminder, the parameters of the smallest crack that can be segmented should be referred to the test image. Moreover, the proposed method is able to identify cracks with a width of only one pixel as analyzed in detail in Section 4.5.1 of the paper. Comment No. 5: In this paper, whether image quality has great influence on the precision of concrete crack segmentation. Response: Thanks to Reviewer for reminder. First, the impact of image quality on segmentation accuracy is certain. We add an analysis of the results in Figure 9 in Section 4.5.1 of the paper, which shows that the present method also performs better in scenes with blurred crack pixels. Comment No. 6: In the introduction, authors should introduc the traditional method to conduct the crack segmentation, In particular, whether the latest clustering method is useful in rock segmentation is introduced, for example: Chen X, Wu H, Lichti D, et al. Extraction of indoor objects based on the exponential function density clustering model[J]. Information Sciences, 2022, 607: 1111-1135. Response: Thanks to Reviewer for reminder, we have supplemented the introduction section with some of the relevant applications of traditional vision, which contains the suggested literature.

Author Response File: Author Response.pdf

Reviewer 2 Report

Manuscript ID: processes-2175989

Title: Pixel-level concrete crack segmentation using pyramidal residual network with omni-dimensional dynamic convolution

Minor revision

This manuscript introduces a pyramidal residual network based on an encoder-decoder using Omni-dimensional dynamic convolution. In summary, the research is exciting and provides valuable results. Still, the current document has several weaknesses that must be strengthened to obtain a documentary result equal to the publication's value.

(1) At the thematic level, the proposal provides an exciting vision, as the automation of crack detection would be a beneficial resource for engineers. Nevertheless, a thorough knowledge of the damage of a built entity is not only limited to deep learning. This issue is an essential limitation of the aspirations of the proposal, whose limitations should be assumed with more rigour and realism in the development of the argumentation of the manuscript.

(2) The document contains a total of 44 employed references, of which 24 are publications produced in the last five years (55%) and 20 in the last 5-10 years (45%), implying a total percentage of 100 % recent references. This way, the total number is sufficient, and their actuality is high. However, vision applications may be mentioned briefly rather than algorithms.

(3) Is it innovative in methods and proves its advanced nature (how is the effect compared with other state-of-art methods)? What are the workload and feasibility/prospects?

(4) The first paragraph introducing the research topic may present a much broad and more comprehensive view of the problems related to your topic with citations to authority references (Seismic Performance Evaluation of Recycled aggregate Concrete-filled Steel tubular Columns with field strain detected via a novel mark-free vision method. Structures, 2022). 

(5) The novelty of the study is not apparent enough. In the introduction section, please highlight the contribution of your work by placing it in context with the work that has been done previously in the same domain.

(6) Generally, the study of the proposed detection techniques is reasonable, and the explanation of the objectives of the work may be valid. However, the limitations of your work are not rigorously assumed and justified.

(7) Vision technology applications in various engineering fields should also be introduced for a full glance at the scope of related areas. For crack classification, references keywords such as integrated generative adversarial networks and improved VGG model, should be mentioned. For crack width detection, references keywords such as backbone double-scale features for improved detection automation, should be mentioned.

(8) In chapter 2, related work: What are the reasons for using SE style attention? What role does it play in the overall network structure?

(9) What is the jump connection strategy? It can be explained a little more clearly.

(10) At the beginning of chapter 3: Methodology, the advantages or functions of this network model are briefly explained.

(11) What are the disadvantages when using the downstream models, and what are the challenges when addressing this problem?

(12) In chapter 4, Experiments and results: Briefly describe the picture sizes of the training and verification sets.

(13) Note that Figure 6 and Figure 5 are the same pictures. Figure 6 is redundant.

(14) In chapter 5, Conclusion: Should mention the scope for further research as well as the implications/application of the study.

(15) I recommend including the limitations regarding the consideration of damage indicated in this review in the limitations assessment. This part of the document can be improved and completed with more rigour.

 

Author Response

Dear Editor and Reviewers: On behalf of my co-authors, we are very grateful to you for giving us an opportunity to revise our manuscript. we appreciate you very much for your positive and constructive comments and suggestions on our manuscript entitled “Pixel-Level Concrete Crack Segmentation Using Pyramidal Residual Network with Omni-Dimensional Dynamic Convolution” (ID: processes-2175989). We have studied reviewers’ comments carefully and tried our best to revise our manuscript according to the comments. The following are the responses and revisions I have made in response to the reviewers' questions and suggestions on an item-by-item basis. Thanks again to the hard work of the editor and reviewer! Comment No. 1: At the thematic level, the proposal provides an exciting vision, as the automation of crack detection would be a beneficial resource for engineers. Nevertheless, a thorough knowledge of the damage of a built entity is not only limited to deep learning. This issue is an essential limitation of the aspirations of the proposal, whose limitations should be assumed with more rigour and realism in the development of the argumentation of the manuscript. Response: Thanks to Reviewer for reminder, we have refined the application of other methods in this field in our presentation and analyze them briefly. Comment No. 2: The document contains a total of 44 employed references, of which 24 are publications produced in the last five years (55%) and 20 in the last 5-10 years (45%), implying a total percentage of 100 % recent references. This way, the total number is sufficient, and their actuality is high. However, vision applications may be mentioned briefly rather than algorithms. Response: Thanks to Reviewer for reminder, we have added applications of traditional vision in related fields to our presentation, which includes earlier approaches. Comment No. 3: Is it innovative in methods and proves its advanced nature (how is the effect compared with other state-of-art methods)? What are the workload and feasibility/prospects? Response: Thanks to Reviewer for reminder, we have done a thorough comparison with other methods of the same type in Section 4.5. The present method can be carried to work on wall climbing robots and UAVs, which is what our team is doing. Comment No. 4: The first paragraph introducing the research topic may present a much broad and more comprehensive view of the problems related to your topic with citations to authority references (Seismic Performance Evaluation of Recycled aggregate Concrete-filled Steel tubular Columns with field strain detected via a novel mark-free vision method. Structures, 2022). Response: Thanks to Reviewer for reminder, we have referenced and cited this literature based on your suggestions. Comment No. 5: The novelty of the study is not apparent enough. In the introduction section, please highlight the contribution of your work by placing it in context with the work that has been done previously in the same domain. Response: Thanks to Reviewer for reminder, we present further work on the method in the introduction section and Section 5, but at present our approach cannot be specified in the related work. Comment No. 6: Generally, the study of the proposed detection techniques is reasonable, and the explanation of the objectives of the work may be valid. However, the limitations of your work are not rigorously assumed and justified. Response: Thanks to Reviewer for reminder, these efforts will be demonstrated in the next step, and we are currently deploying the method to the Nvidia Jetson device. The wall climbing robot carries the module for subsequent work rather than computer simulation. Comment No. 7: Vision technology applications in various engineering fields should also be introduced for a full glance at the scope of related areas. For crack classification, references keywords such as integrated generative adversarial networks and improved VGG model, should be mentioned. For crack width detection, references keywords such as backbone double-scale features for improved detection automation, should be mentioned. Response: Thanks to Reviewer for reminder, we have added vision applications in some fields in the introduction section and briefly listed the approaches based on vision applications with the suggested keywords. Comment No. 8: In chapter 2, related work: What are the reasons for using SE style attention? What role does it play in the overall network structure? Response: Thanks to Reviewer for reminder, we correct the statement about SE in Section 2.1.2 and describe in detail the role of modules in the network. Comment No. 9: What is the jump connection strategy? It can be explained a little more clearly. Response: Thanks to Reviewer for reminder, we correct the statement about Skip Connections (not jump connect) in Section 2.1.1 and describe in detail how it is implemented. Comment No. 10: At the beginning of chapter 3: Methodology, the advantages or functions of this network model are briefly explained. Response: Thanks to Reviewer for reminder. At the beginning of chapter 3, we add the advantages of small model and high accuracy of this method. Comment No. 11: What are the disadvantages when using the downstream models, and what are the challenges when addressing this problem? Response: Thanks to Reviewer for reminder. The dataset may not be comprehensive enough when solving downstream models. The solution is to continue to expand the dataset subsequently, especially for the specific objects actually detected. Comment No. 12: In chapter 4, Experiments and results: Briefly describe the picture sizes of the training and verification sets. Response: Thanks to Reviewer for reminder, we have added a description of the experimental image size in Section 4.1. Comment No. 13: Note that Figure 6 and Figure 5 are the same pictures. Figure 6 is redundant. Response: Thanks to Reviewer for reminder, Figure 6 is mainly intended to illustrate the experimental parameters for Table 3. To differentiate, we have modified Figure 5 accordingly. Comment No. 14: In chapter 5, Conclusion: Should mention the scope for further research as well as the implications/application of the study. Response: Thanks to Reviewer for reminder, we have added further work at the end of Chapter 5. Comment No. 15: I recommend including the limitations regarding the consideration of damage indicated in this review in the limitations assessment. This part of the document can be improved and completed with more rigour. Response: Thanks to Reviewer for reminder, we have made improvements and refinements based on your suggestions. Thank you again for your suggestions.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

All the comments have done

Author Response

Dear Editor and Reviewers:

On behalf of my co-authors, we are very grateful to you again for giving us an opportunity to revise our manuscript.

 

Reviewer 2 Report

The authors have successfully addressed all my comments. Therefore, I recommend the publication of this manuscript. However, some citations are missing (Lines 67-71):

For example, researchers can augment digital image data with Generative Adversarial Networks (GAN) and combine them with improved Visual Geometry Group (VGG) networks to achieve crack classification. In terms of crack width measurement, a new crack width measurement method based on backbone dual-scale features can improve detection automation.   I suggest adding corresponding articles, in the first sentence, for instance:

Automatic classification of asphalt pavement cracks using a novel integrated generative adversarial networks and improved VGG model. Engineering Structures 2023.

     

Author Response

Dear Editor and Reviewers:

On behalf of my co-authors, we are very grateful to you for giving us an opportunity to revise our manuscript. we appreciate you very much for your positive and constructive comments and suggestions on our manuscript entitled “Pixel-Level Concrete Crack Segmentation Using Pyramidal Residual Network with Omni-Dimensional Dynamic Convolution” (ID: processes-2175989).

We have added the missing references as follows (Lines 56-60):

For example, researchers can augment digital image data with Generative Adversarial Networks (GAN) and combine them with improved Visual Geometry Group (VGG) networks to achieve crack classification [14]. In terms of crack width measurement, a new crack width measurement method based on backbone dual-scale features can improve detection automation [15].

14. Que, Y.; Dai, Y.; Ji, X.; Kwan Leung, A.; Chen, Z.; Jiang, Z.; Tang, Y., Automatic classification of asphalt pavement cracks using a novel integrated generative adversarial networks and improved VGG model. Engineering Structures 2023, 277, 115406.
15. Tang, Y.; Huang, Z.; Chen, Z.; Chen, M.; Zhou, H.; Zhang, H.; Sun, J., Novel visual crack width measurement based on backbone double-scale features for improved detection automation. Engineering Structures 2023, 274, 115158.

 

Round 3

Reviewer 2 Report

I recommend the publication of this manuscript since the authors have successfully addressed all the comments.