MA-Xnet: Mobile-Attention X-Network for Crack Detection

Round 1

Reviewer 1 Report

I don't feel this work is sufficient for this journal. Simply changes layers wont make good sense. Kindly implement many applications with your proposed work. 

Kindly do many pretrain applications with your network and compare existing algorithms.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The received paper compares different deep learning techniques for crack segmentation. Conventional methods such as U-Net have a lot of parameters that make the training difficult. Alternatively, MA-XNET is designed for crack segmentation that leads to higher performance while it has fewer parameters rather other segmentation methods. The paper is well written, however, some minor modifications are required

- The main focus of the abstract is using different deep learning techniques for segmentation in general. Since the topic of the paper is using segmentation for crack detection, the abstract should be revised in this regard.

- Typo in line 51.

- Line 188, a blank is needed between “Sji” and “measures”.

- The locations of numbers of all formulas are not suited. Need to be revised.

- A separate section for describing the datasets used in this study is necessary.

- In addition to the number of parameters, it would be great to bring the time of training and testing.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The authors proposed a new approach to perform the assessment of cracks based on the semantic segmentation model. This is an important topic for monitoring systems and the article can be accepted for publication after the authors answer the following questions:

1 – in the introduction section the authors attested that “However, these algorithms are generally vulnerable to noise and lighting, which affect 30 the detection results”. I understand that the authors are mentioning the algorithms but some techniques like crack detection by vibrothermography like in the articles:

https://www.sciencedirect.com/science/article/abs/pii/S0963869517304097

https://www.sciencedirect.com/science/article/abs/pii/S0963869518303980 

are not sensitive to temperature variations and depend on imaging processing.

Please improve the discussion about the monitoring field. This is important.

2 – Improve the explanation of equation 11.

3 – why the limit in Eq12, is 6? Please, give a theoretical analysis.

4 – The legend of figure 7 is not close to the figure.

5 – how this algorithm can detect the crack size evolution?

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

No Comments

Reviewer 3 Report

The authors improved the article and it can be accepted for publication.