Review Reports
by
- Daniel Ochs 1,*,
- Karsten Wiertz 2 and
- Devendra Singh Dhami 3,6
- et al.
Reviewer 1: Anonymous Reviewer 2: Chaitanya Pande Reviewer 3: Anonymous Reviewer 4: Anonymous
Round 1
Reviewer 1 Report (Previous Reviewer 1)
Comments and Suggestions for AuthorsThe author made significant revisions to the article, but I still believe that this work lacks a certain degree of innovation. Besides, compared to other state-of-the-art CD frameworks, there exist significant gaps for the proposed method in Precision as shown in Table 1 and Table 2. In my opinion, there is some way to go before this paper gets published.
Comments on the Quality of English LanguageMinor editing of English language required.
Author Response
Dear Reviewer 1,
We thank you for your time and the valuable and positive feedback, which helped us improve the quality of our work.
As argued before, we believe our work has enough innovation and is valuable for the community. Additionally, we acknowledge that, indeed, the ChangeFormer has slightly less precision but performs better in the recall metric, which is more valuable for pipeline monitoring. This point is also mentioned in the paper; we want to emphasize again that DTCDSCN and TinyCD could be alternatives to the ChangeFormer, which we discuss in lines 312-319 in the revised manuscript.
Reviewer 2 Report (Previous Reviewer 2)
Comments and Suggestions for Authorsaccepted
Author Response
Dear Reviewer 2,
We thank you for your time and the valuable and positive feedback, which helped us improve the quality of our work.
Reviewer 3 Report (Previous Reviewer 3)
Comments and Suggestions for AuthorsThe revised paper is well-written and provides enough details. I recommend accepting.
Author Response
Dear Reviewer 3,
We thank you for your time and the valuable and positive feedback, which helped us improve the quality of our work.
Reviewer 4 Report (New Reviewer)
Comments and Suggestions for AuthorsDear authors you can find my report in the PDF file
Comments for author File: 
Comments.pdf
Author Response
Dear Reviewer 4,
We thank you for your time and valuable feedback, which helped us improve the quality of our work. We added your suggestions to the manuscript and highlighted them in blue. We changed the sentence in line 192 to be more precise and put table captions above the table.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report (Previous Reviewer 1)
Comments and Suggestions for AuthorsThe author has further improved the suggested comments, which meet the publishing requirements.
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis paper proposes to use transformers trained with low-resolution images in a change detection setting to detect pipeline risks. This paper is quite interesting, but the authors should kindly consider some changes and comments for potential improvement.
1. The abstract needs to be rewritten to describe as much of your work as possible rather than the research background.
2. The summary of related work is not complete enough and needs to be supplemented.
3. Regarding the CD framework, there is a lack of innovative presentation that merely presents a simplified version of the original method (Bandara, 2022). Bandara, W.G.C.; Patel, V .M. A Transformer-Based Siamese Network for Change Detection. In Proceedings of the IEEE International Geoscience and Remote Sensing Symposium, 2022, pp. 207–210.
4. Is the proposed framework universal for multispectral or hyperspectral remote images that can provide rich feature information?
5. Please compare with the current state-of-the-art CD frameworks.
6. If the CD framework can detect types of changes, it is more worthy of promotion.
Comments on the Quality of English LanguageMinor editing of English language required.
Reviewer 2 Report
Comments and Suggestions for Authors1. Author should add the results in the Abstract section , total revised this abstract.
2. Author should improved the review of papers in the Introduction, this sections is showing some part is a very poor.
3. No study area information mentioned in the paper.
4. Aim and objective must add in the paper with problem statement add in the paper.
5. Results and Discussion part must improve and author should add novelty of paper.
6. Discussion section separately write and add novelty. conclusion is not show any novelty of work
Reviewer 3 Report
Comments and Suggestions for AuthorsThe paper presents a transformer-based change detection methods for pipeline inspection. The paper lacks the technical strengths. What is the role of \alpha in eq 2. The comparison with other state of the art is not sufficient. How neural network is trained is also not explained in detail. The list of risks needs to list in pipeline inspection. The description algorithms is not enough. How is data labelled? Whether time series data will be more useful than only two images? how alpha is helping in recall. Technical description needs to improve. add pseudo code. of the proposed scheme. A fair comparison with other methods is needed.
Reviewer 4 Report
Comments and Suggestions for AuthorsThe paper presents a change detection (CD) procedure applied to satellite images collected along the track of gas pipelines: significant changes mean a possible risk for the pipeline. The procedure is based on transformers trained with low-resolution images (3m): the authors show an example with 1372 supervised cases (80% training, 10% validation, 10% testing), that obtains better results wrt similar methods from the literature. The reading is fluent and the topic is interesting.
Although keeping in mind this is presented as a communication (1), at the end I got the following "feelings"
- a scientific paper should contain a validated and repeteable proof of concept, while this one seems a case history with a not fully documented tailored solution, more suitable for a conference
- the 1372 cases contain also "false positives"? If yes, in which percentage? If not, how can the procedure learn about them?
- please show a case where the procedure fails (missed alarm)
- images are in color: how this is managed?
- please give more details on the selected features and on the deep learning strategy
In conclusion, I think this manuscript does not have enough scientific material and rigor to be published in Remote Sensing.
----------------------------------
(1) from Remote Sensing web site: "Communications are short, rapid articles that report results of important preliminary research findings, ongoing research projects or results of limited significance. Short articles might include the discovery or development of new materials, or cutting-edge experiments and theories."
Comments on the Quality of English Languageenglish is ok