Intelligent Detection Method for Concrete Dam Surface Cracks Based on Two-Stage Transfer Learning

Round 1

Reviewer 1 Report

Please check the attached file. 

Comments for author File: Comments.pdf

Author Response

Dear reviewers:

For my coauthors, we thank you very much for your careful and thorough reading of this manuscript and for your thoughtful comments and constructive suggestions, which have been very helpful in improving the quality of the manuscript (water-2395285). We have carefully studied your comments and made corresponding corrections. Words in red track the changes made in the manuscript, and the responses are presented in blue text. We hope these revisions meet with your approval. We answer your questions or comments in detail in the respond_letter.

Author Response File: Author Response.pdf

Reviewer 2 Report

-Novelty of the paper should be highlighted. There are so many papers on same topic. it is difficult to see what is new.

-Is the dam crack database shared? Authors generally tend to avoid this question and write shared on request. But eventually dont reply to emails while asking for queries. This should be clear as this is important for replication of results. 

-About the database it is not clear? is it auhors or shared database? location and how it was gathered?

-Fig 1 The first row shows images of cracks in the open dataset, reference should be there of open dataset.

-More methods should be summarised for dam crack assessment .IN that essense, paper is weak.

-In related works section; some reference of DL can be added. Two-stage method based on the you only look once framework and image segmentation for crack detection in concrete structures

Major revisions are needed.

 

Minor improvements. 

 

Author Response

Dear reviewers:

For my coauthors, we thank you very much for your careful and thorough reading of this manuscript and for your thoughtful comments and constructive suggestions, which have been very helpful in improving the quality of the manuscript (water-2395285). We carefully studied your comments and made corresponding corrections. Words in red are the tracked changes we made in the manuscript, and the responses are presented in blue text. We hope the changes meet with your approval. We answered your questions or comments in detail in the respond_letter.

Author Response File: Author Response.pdf

Reviewer 3 Report

This manuscript introduces an intelligent detection method for concrete dam surface cracks based on two-stage transfer learning. The ResNet50 network is used as the backbone feature extraction network and a Multilayer Parallel Residual Attention (MPR) is integrated into its jump connection path to improve the focus on critical information for clearer fracture edge segmentation. In general, the proposed method has better performance than the original UNet network, but I still think that the study is not innovative enough.

General considerations:

（1）       At the thematic level，this study investigates the application of deep learning-based semantic segmentation networks in dam crack detection. And the solution idea is proposed for the problems of data set scarcity and network structure. However, I think the article still needs to highlight the main technical contributions.

（2）       The document contains a total of 58 employed references, of which 39 are publications produced in the last 5 years (67.2%), 19 in the last 5-10 years (32.8%). In this way, the total number is sufficient, and their actuality is high. It is suggested to add some literature on the application of vision in different fields such as Novel visual crack width measurement based on backbone double-scale features for improved detection automation, Engineering Structures.

（3）       Technique concerns: The modification of the network in this study is not considered a prominent technical contribution. Although migration learning is applied in this study, there is no essential innovative work.

Title, Abstract and Keywords:

（4）       The abstract is complete and well-structured and explains the contents of the document very well. Nonetheless, the part relating to the results could provide numerical indicators obtained in the research.

Chapter 1: Introduction

（5）       The literature review on deep learning & vision algorithms is not detailed enough, for example, there are many algorithms for target detection that are also used in this field but are not discussed here. The authors do not make it clear why semantic segmentation is more applicable to this study.

（6）       The study of the dataset also needs to be discussed here and summarize the criteria for collecting the dataset later.

Chapter 2: Related work

（7）       I think the authors need to enumerate the performance metrics of various semantic segmentation networks and analyze the advantages and disadvantages of each.

Chapter 3: Proposed approach

（8）       If MPR is an original work, please give more details about it. If the MPR is simply an application of someone else's technique, please cite the relevant literature.

Chapter 4: Experiments and evaluation indicators

（9）       In this paper, it is emphasized several times that the algorithm can be used for the data collected by UVA, but UVA is not involved in the data and equipment of the experiment.

Chapter 5: Conclusions

 

（10）    Reviewing the entire study, I still could not find any outstanding original work among them.

Comments for author File: Comments.docx

none.

Author Response

Dear reviewers:

For my coauthors, we thank you very much for your careful and thorough reading of this manuscript and for your thoughtful comments and constructive suggestions, which were very helpful in improving the quality of the manuscript (water-2395285). We carefully studied your comments and made corresponding corrections. Words in red are the tracked changes in the manuscript, and the responses are presented in blue text. We hope the changes meet with your approval. We answered your questions or comments in detail in the respond_letter.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have partially addressed my review queries. I will direct the following questions to the previous round of queries for reference: 

1. Previous query 2: The authors have partially discussed the use of singular spectrum analysis for road surface detection. However, they have completely missed out on the aspect of eigen perturbation. Real-time singular spectrum analysis using eigen perturbation has found a lot of applications in road-surface assessment that should be discussed in the manuscript. 

2. Previous query 6: This query has not been adequately addressed. Nowhere in the query has there been a mention of a drone. This query should be addressed based on the effectiveness of the proposed approach and compared against eigen perturbation based real-time single-sensor monitoring - which is primarily missing from the context. Please address this once again with a fresh perspective. 

Author Response

Dear reviewers:

For my coauthors, we thank you very much for your careful and thorough reading of this manuscript and for your thoughtful comments and constructive suggestions, which have been very helpful in improving the quality of the manuscript (water-2395285). We carefully studied your comments and made corresponding corrections, and have provided further responses to the questions we were unable to fully answer last time. Words in red are the tracked changes we made in the manuscript, and the responses are presented in blue text. We hope the changes meet with your approval. We answered your questions or comments in detail in the respond_letter.

Author Response File: Author Response.docx

Reviewer 3 Report

accept

Author Response

Dear reviewers:

Thank you again for your comments on our manuscript entitled "Intelligent detection method for concrete dam surface cracks based on two-stage transfer learning" (water-2395285). The comments were valuable and helpful. Thank you for your time and consideration.