Marine Oil Spill Detection from Low-Quality SAR Remote Sensing Images

Round 1

Reviewer 1 Report

In this manuscript, the authors propose novel methods to perform automatic oil spill identification from low-quality SAR images; The proposed methods are compared against state-of-the-art NN-based solutions (e.g., Unet and  DeepLabV3+). Moreover, the authors point out some areas of improvement concerning the Deep SAR Oil Spill (SOS) dataset and propose a method to reduce the noise in SAR images by means of the FFDNet.

Major remarks:

1. Equation 16 (MIoU) is wrong. Is this equation used to compute the MIoU results reported in subsections 6.2 and 6.3? Please update equation 16 and ensure the numerical results in subsections 6.2 and 6.3 are correct. A correct equation for MIoU can be found in subsection IV.B of the paper referred to as [14] in your manuscript.

2. The used loss function is described in subsection 3.3, while other training parameters are described at the beginning of section 6 (lines 461 to 467). All the training parameters should be in the same section. And, the reference to the training at line 403 should be updated accordingly.

3. In line 579 the authors claim that the proposed TransUNet-based method is interpretable. The author should motivate this claim more precisely.

4. Since the authors make a comparison against the work of Qiqi Zhu et al., they should add a few lines in subsection 6.6 describing why some results differ (e.g., why Qiqi Zhu et al. report higher MIoU for U-Net, etc.).

 

Minor remarks:

1. In section 6.5 the authors report the training performance but no inference performance is reported (inference time, memory used during the inference, etc.). Nowadays knowing the inference performance and the selected target hardware to deploy the developed system is of paramount importance. The authors should add this information and compare the performance against the other works. Usually, high-performance methods require more hardware resources and have better accuracy and MIoU when compared to real-time solutions. Such a comparison is mandatory to understand what is the application scenario foreseen by the authors. Some examples of real-time solutions for oil spills/ships identification in SAR images are:

  1.1. https://www.mdpi.com/2072-4292/13/18/3606

  1.2. https://ieeexplore.ieee.org/abstract/document/9739778

 

2. In line 232, it would  be nice to have the link to the corrected SOS dataset in the final version of this manuscript so the reader will be able to directly access it;

3. In line 237, I think "sample" should be "train";

4. In line 484. M1, M2, and M3 should be clearly defined (e.g., "M3 is the Multi-model Ensemble", etc.);

5. In line 596, "its" should be removed;

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

GENERAL COMMENTS

This manuscript aims to detect and correct SOS dataset errors, improve image segmentation performance, and provide research guidance in the field of marine oil spill detection with low-quality SAR remote sensing images.

Basically, it is about extending current methodologies already applied in other domains to improve the current capabilities for detecting oil spills at sea. Therefore, although this manuscript does not provide anything significantly relevant in methodological terms, the suggested multi-faceted solution for oil spill detection based on the SOS dataset and the proposed models show better detection performance compared to conventional semantic segmentation models.

The state-of-art regarding oil spill detection after it has occurred is acceptable, as SAR remote sensing methods are generally the most suitable. However, in estimating the drift direction of the oil spill and the weathering processes that affect its characteristics, especially under adverse weather conditions, numerical methods are of paramount importance.

Therefore, although this manuscript has the potential for possible publication, it has some limitations in conceptual terms. After the occurrence and detection of an oil spill at sea, it is important to implement measures to mitigate the negative impacts associated with oil spills, and for that modeling tools are of paramount importance in the forecasting of oil slick evolution.

In fact, mathematical modeling is a very powerful tool for managing an oil spill accident, namely to monitor the evolution of the oil slick taking into account the spreading and weathering processes, such as evaporation, vertical dispersion, emulsification, and viscosity changes, and also for determining preventive measures.

It should be noted that numerical models together with GIS technology make it possible to reorient the evolution of the oil slick at sea and redefine its characteristics, such as area and thickness, at detection points.

Regarding the content, please note that Section 3 and most of Section 4 may be abbreviated as they are primarily replicas of the literature and are possibly less interesting for JMSE readers. On the experimental results and analyses shown in section 6, they seem to be promising.

Keeping the core of the manuscript, my suggestion is that essentially the Introduction and the Discussion be reworked with additional material on oil spills at sea in the various stages, namely the detection of the occurrence, the evolution of the oil slick (taking into account the spreading and weathering processes) and the possible measures that should be implemented to mitigate the negative consequences. Some possible supporting references are:

- Pinho et al., 2002. Numerical modelling of oil spills in coastal zones. A case study. Water Studies, Volume 11, Pages 35-45. First International Conference on Oil and Hydrocarbon Spills: Modelling, Analysis and Control, Oil Spill III, Code 63109.

- Inan, 2011. Modeling of Oil Pollution in Derince Harbor. Journal of Coastal Research, SI 64 (Proceedings of the 11th International Coastal Symposium), 894–898. Szczecin, Poland, ISSN 0749-0208. 

- Cho et al., 2012. Numerical Simulation of Oil Spill in Ocean. Journal of Applied Mathematics, Volume 2012, Article ID 681585, doi:10.1155/2012/681585.

- Iouzzi et al., 2023. Modeling of the Fate and Behaviors of an Oil Spill in the Azemmour River Estuary in Morocco. Water, 15, 1776. https://doi.org/10.3390/w15091776.

Many other flaws exist throughout the manuscript requiring moderate revision, including additional material.

Looking to be constructive, in addition to the General Comments above, I would like to draw attention to the following Specific Remarks.

SPECIFIC REMARKS

- Avoid using keywords that already exist in the title, such as "marine oil spill detection" and "low-quality remote sensing images". 

- Avoid repetitions in the Abstract and throughout the manuscript; just an example, "this paper" is repeated 6 times in the Abstract.

- Abbreviations should be defined at first mention and used consistently thereafter.

- Lines 68-70 read "It is estimated that approximately ... for 5% of total offshore oil production". A reference should be provided.

- A reference or references for the several accidents reported in lines 74-78 should be provided.

- Consider references in the content of lines 173-178.

- The purposes of Figures 1 and 2 are not clear. Each figure should have a concise caption describing accurately what the figure depicts. 

- Figure 3 is adapted from [27]. This should appear in the figure caption.

- As for the captions of Figures 1 and 2, also the captions of Figures 3 and 4 should be more explanatory of the content of the figures. Note that all figures, together with their captions, must be self-explanatory.

- Line 347 reads "...in the middle part of Figure 5"; should be "...in the middle part of Figure 4". 

- Line 365 reads "According to the Goodmann model..."; consider a reference to this model or more explanatory material.

- Line 417 reads "...use the or operation to obtain..." and line 421 reads "...denotes the logical or operator"; rework these contents.

- Line 450 reads "The web server uses ajax technology to start polling until it finds..."; elaborate on the "ajax technology".

- Consider reworking the current Discussion section with more comprehensive material and a new Conclusions section.

As a final point, note the various grammatical and phrasing inaccuracies throughout the manuscript and other occasional errors.

The English language is moderate. All parts of the manuscript must be carefully checked, because there are some linguistic mistakes that should be fixed. Authors should request the help of a native English-speaking proofreader.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

 

The paper provides a concise and clear overview of the contributions made by the paper in marine oil spill detection using low-quality SAR images. The paper addresses several key aspects, but there are some minor and major flows:

 

1.       Dataset Analysis and Corrections: The paper conducts a detailed analysis of the Deep SAR Oil Spill dataset (SOS dataset) and identifies problems in the original dataset. How it corrects the errors and makes improvements to ensure the dataset's reliability for marine oil spill detection research. Need justification!

2.       Reproduction of Experiments: The paper identifies errors in the experimental results published in the original literature with the SOS dataset. How did the authors reproduce the experiments? Have they used any new measurement techniques?

3.       Proposed Deep Learning Methods: The paper proposes three deep learning-based methods for marine oil spill detection: a direct detection method based on Transformer and UNet, a method using FFDNet and TransUNet with denoising before detection, and an integrated multi-model learning approach. The performance advantages of these methods are validated by comparisons with other semantic segmentation models like UNet, SegNet, and DeepLabV3+, but the novelty is not mentioned; what the authors have done with a transformer to improve the results.

4.       Extension of literature will be appreciated “https://doi.org/10.1016/j.knosys.2023.110525” and cite behind the DL and ML to enrich the literature.

5.       System Architecture: The paper presents a feasible, highly robust, and easily scalable system architecture approach that addresses practical engineering applications for marine oil spill detection. How the authors claimed a robust is they proposed architecture, model, or system. Need clarifications and mention the advantages and limitations as well!

 

 

moderate 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Overall, the manuscript has improved slightly in the review process. Some text rearrangements were performed, and pertinent clarifications were introduced. Some unnecessary material has been removed, including tables, and other more appropriate content has been added.

However, not all issues were properly addressed. Some are of minor importance, but others are relevant and should be considered.

The captions of Figures 1 and 2 repeat the captions of all subplots shown in each figure, why? Also, note that the caption of Figure 1 is repeated on lines 194-196. On the other hand, the captions of Figures 1 and 2 do not provide enough information to guide the reader about what the figures are intended to depict. Therefore, both captions must be rethought.

Line 452 reads "Figure 7. The information processing flow"; should be "Figure 8. The information processing flow".

The sentence "The values of each element of the confusion matrix are shown in Table 3, and they are the basis of the data in Table 2, which is used to prove the authenticity of each performance indicator in Table 2" (lines 504-506) is a bit unclear. Consider rephrasing this sentence and possibly exchanging Tables 2 and 3.

Also, the sentence "The experimental results are shown in Table 4, and the confusion matrix data on which they are based are shown in Table 5" (lines 511-513) should be reworked and, likewise, Tables 4 and 5 must also be exchanged.

Lines 592-593 read "In order to ensure the reliability of the experimental data, this manuscript includes Table 3 and Table 4 to display the raw confusion matrix data..."; aren't these the current Tables 3 and 5?

Consider moving the sentence "In subsequent research, we will attempt to utilize model compression techniques to reduce resource consumption in practical applications" (lines 567-569) to the end of the current 7. Discussion section. Also, consider heading this section as "Discussion and Conclusions".

Carefully check the references. Not all are spelled correctly; just two examples:

- Reference 20 should be: Pinho J.L.S.; Antunes Do Carmo J.S.; Vieira J.M.P. Numerical modelling of oil spills in coastal zones. A case study. First International Conference on Oil and Hydrocarbon Spills: Modelling, Analysis and Control, Oil Spill III. Water Studies, Volume 11, 2002, pp. 35-45, https://www.witpress.com/Secure/elibrary/papers/OIL02/OIL02004FU.pdf

- Reference 21 should be: Inan A. Modeling of Oil Pollution in Derince Harbor. Proceedings of the 11th International Coastal Symposium, Poland. Journal of Coastal Research, SI 64. 2011, 894–898, file:///C:/FCTUC/Reviews/JMSE_MDPI_Reviews_31.07.2023/JMSE_Review_58_1st/Modeling%20of%20Oil%20Pollution%20in%20Derince%20Harbor.pdf

Lines 603-605 are an excellent compromise.

Overall, the English language and style are acceptable; anyway, a minor spell check is advised, especially on punctuation, some grammatical constructions, and other occasional errors.  

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The Authors successfully addressed my comments and recommendations. Good Luck!

minor editing is required!!

Author Response

I have diligently revised the manuscript, making detailed improvements. I apologize for any oversights on my part and sincerely thank you for your patient guidance.