A Novel Shipyard Production State Monitoring Method Based on Satellite Remote Sensing Images
Round 1
Reviewer 1 Report
In this paper, authors propose the use of satellite remote sensing data to monitor shipyard production. The use of RS data for this scope is novel.
The authors propose the use of transfer learning CNNs in order to extract semantic information from the remote sensing images. Moreover, they introduce a novel evidence fusion method.
Authors demonstrated that the proposed method works well in their two experiments.
This paper is interesting and suitable for publication.
The paper is clearly written and well organised.
Experimental results are convincing.
Overall, it is a good work in my view and worth publishing as a paper.
I have the following comments:
- The complexity of the used CNNs should be added.
- Additional details on the training set of CNNs should be included.
- The training optimizer should also be included.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The research manuscript entitled “A novel shipyard production state monitoring method based on satellite remote sensing images,” is well-organized along with adequate details and descriptions. The authors developed a new method to monitor shipyard production states using multi-temporal satellite remote sensing data. They employed transfer learning convolutional neural networks (CNNs) to extract high-level semantic information from optical remote sensing images. To address conflicts in evidence from different core shipyard sites, they introduced correlation and similarity metrics. They applied and tested their method in the Yangtze River Delta and the Bohai Sea, demonstrating that their approach more accurately determines shipyard production states compared to traditional methods. The results shown in the manuscript are promising. However, there are few comments that needs to be addressed before it can be recommended for publication in the journal of Remote Sensing, MDPI.
1. The authors mention the use of transfer learning convolutional neural networks (CNNs), but it would be helpful to specify which pre-trained networks were used and why they were chosen.
2. The conclusion seems to have repeated the main content from the introduction. A conclusion should provide a succinct summary of the findings and the implications of the study, rather than repeating the methodologies used. It might be more beneficial for the authors to discuss the broader impact of their work, potential future applications, and the next steps in their research.
3. The manuscript would benefit from a more detailed description of how the accuracy of their proposed method was quantified, especially in comparison to traditional DS evidence fusion methods. What metrics were used, and how did they quantify performance in different scenarios?
4. The authors mention that their method is superior to existing methods, but it would be helpful to see a comparative analysis with other leading methods in the field. This could provide a clearer picture of how their method stands out.
5. Every methodology has limitations. It would add depth to the manuscript if the authors discussed potential limitations of their method and any challenges they faced.
6. It is important for authors to ensure that terminology and acronyms are consistently used throughout the manuscript. For instance, remote sensing (RS) and high-resolution RS (HRS) are introduced, but it would be worth checking if these are consistently applied.
7. Ensure all the cited references are relevant and current. It might be worth suggesting the inclusion of any recent seminal works in the field that the authors might have missed.
A thorough grammar and spell check should be conducted to ensure the manuscript is free of errors.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
1. In lines 65-69, you claim that the decision-level fusion is more accurate than the pixel or feature-level fusion. However, no relevant experiments have been conducted for proof.
2. There are many handcrafted steps and rules for the decision-level fusion. I was wondering if the method is robust. A cross-experiment on the Yangtze River Delta and the Bohai Rim is recommended.
3. What is the meaning of "the ith SCS in optical data" in line 168? Does it mean the images captured at different times in the same spot?
4. There is no comparison to other methods. If there is no similar method on this topic, all possible methods should be listed and compared.
5. The images in some Tables, e.g. Tables 3, 8, and 12 are not aligned well.
There are several writing issues in the paper:
1. There should be a space between two sentences, e.g. lines 13, 36, etc.
2. There are grammar problems in some sentences, e.g. lines 124-126.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
The authors have improved the structure and clarity of their study titled “A novel shipyard production state monitoring method based on satellite remote sensing images.” They have also tackled the required technical specifics. However, the reviewer believes that the manuscript needs few minor revisions before it can be accepted for publication in the journal of Remote Sensing, MDPI.
1. While the authors elaborated on why simpler convolutional neural networks were chosen, the exact models used still remain unspecified. As explained in the lines 161-167 of the revised manuscript, please provide more explanations and details of the specific pre-trained networks utilized and clarify if custom architectures were developed.
2. The acknowledgment of limitations is essential, however, it would be insightful to have a discussion on potential solutions or workarounds for these limitations, even if they might be part of future work.
3. It is commendable that a detailed description was added concerning the accuracy quantification. However, consider offering direct comparisons, if possible, perhaps in the form of percentage improvements, between your method and the traditional DS evidence fusion methods.
4. The inclusion of recent literature is a step in the right direction. Ensure all new references are adequately integrated into the manuscript's narrative and not just listed. In addition, please consider the following recently published papers in RS journal, to help readers better understand the rationale behind the topic and different monitoring schemes using satellite images and machine learning algorithms: https://doi.org/10.3390/rs15123095, https://doi.org/10.3390/rs14010201.
5. The conclusion has been revised, and it sounds much better than before. Please ensure it not only highlights the study's achievements but also points towards future directions or potential real-world implications.
6. While improvements in the language quality are noted, it would be beneficial to undergo one more round of proofreading to ensure clarity and coherence as there are still some grammatical errors in the manuscript, especially in the revised sections.
While improvements in the language quality are noted, it would be beneficial to undergo one more round of proofreading to ensure clarity and coherence as there are still some grammatical errors in the manuscript, especially in the revised sections.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
All my concerns have been replied to, except for the images in Table 3. Thanks for your effort.
Please carefully check the writing again.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
