Oil Spill Identification in Radar Images Using a Soft Attention Segmentation Model

Round 1

Reviewer 1 Report

The scope of the paper is to present a numerical method to identify if an oil slick occurs on a radar image of the sea. The authors discuss of four segmentation models (Table 4) and six detection models (Table 3). The radar images enable from several institutions (line 180-182) including the satellite ERS. The theoretical part of the paper is the section 2. The clearness of the paper appears insufficient, and several examples will follow. It results the recommendation to ask for a large revision of the paper to give more details and explanations of the research work done. First example concerns “multi-scale” (line 8) which could have many significations, including multi-frequency, multi-spatial (resolution), multi-spectral. The reader could aim to have more details on this aspect given in the paper abstract. Second example is the authors’ usages of ERS satellite images. It existed two ERS: ERS-1 and ERS-2. A new generation is Sentinel. Here, the reader could ask for a precise situation of the research presented. We give here after some suggestions from the scientific and editorial points of view.

Science

The authors use the X-band, but others bands are available, the L-band and the C-band which is close to the X-band. This hypothesis could be detailed.

The convolution coefficient dimensions 1*1*512 and 3*3*256 could be justified. The paragraph L105-L116 on the method appears unclear, idem for the part between the figures 2 and 3.

Section 2.2 contains more illustrations but it remains obscure. At contrary, equations (6) to (9) give some clearness. The section after figure (9) could be more comprehensible.

Edition

Sentence line 7-9 is unclear (abstract).

Occurrence of the word feature is 60. It may be reduced, to improve the paper flow.

]proposed, line 50.

Check “Relu” line 145.

Plural figures 4 and 5 line 150.

With s , represents, caption fig (5).

Is “i” related to H, W or C in equation (3). It may be explained, as well as the summation.

L_mask depends on (i) in eq(4), and eq(3). It could be explicit.

L_IoU depends on (s) in eq(5). It could be explicit.

ERS-1 mission was 1991-2000 and ERS-2 from 1995 to 2011. The authors could give the spatial resolution in the images proposed (6 meters?).

Fig(6) concerns the Dalian region and the author could present this main point.

Is N_ii concerned by Table 2? A definition could be given for these numbers.

Plural indicators, line 215

“Attention U-net”, line 253

First letters: Marine Pollution Bulletin, refs 1 and 17, Sensors, ref 5.

X-band, ref 27

Author Response

Thank you for relevant and useful comments and suggestions, which are of great help to my thesis revision. For detailed replies, please see the attached PDF document.

Author Response File: Author Response.pdf

Reviewer 2 Report

The manuscript is well-written and sufficient explanation of the method has been provided. In recent years, there is an increasing number of publications relating to remote sensing for oil spill detection available in vast number of journals, the method described by the authors displayed some uniqueness.  I think the manuscript can be accepted in it's form. No major English grammatical or spelling errors were observed.

Author Response

Thank you for your review. It helped me a lot.

Reviewer 3 Report

Paper is good only minor changes needed

Comments for author File: Comments.pdf

Author Response

Thank you for relevant and useful comments and suggestions, which are of great help to my thesis revision. We have made all the changes.

Author Response File: Author Response.pdf