Scattering-Point-Guided RPN for Oriented Ship Detection in SAR Images
Round 1
Reviewer 1 Report
The paper proposes a novel oriented ship detection method in SAR images named SGOD for detecting under different imaging conditions. They propose an riented two-stage detection module based on the scattering characteristics. Scattering-point-guided region proposal network (RPN) is used for predicting possible key scattering points and increase attention to the vicinity of key scattering points and reduce attention to background and noise. At last, supervised contrastive learning is introduced to alleviate the problem of minute discrepancies among SAR object classes. Extensive experiments are conducted on the dataset from to demonstrate the effectiveness of SGOD.
On the whole, the paper is well organized and innovative. The experimental arrangement of the paper is reasonable and the experimental results are credible. However, the following problems exist.
1.English and grammar need to be improved
2.The number of current related references is insufficient. The authors should add the following work, as it constructed a oriented SAR ship detection dataset and share to peers:
XU Congan, SU Hang, LI Jianwei, et al. RSDD-SAR：Rotated Ship Detection Dataset in SAR Images [J]. Journal of Radars, in press. doi: 10.12000/JR22007.
He Yi-shan, Gao Feo, Wang Jun et al.. Learning polar encodings for arbitrary-oriented ship detection in SAR images[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021, 14: 3846-3859.
3.The related works should introduce the papers those are related to your manuscript. The now related works are similar to introduction which is not suitable.
4.Section 6 is too brief, it should be more detail.
5.The number of samples of each class in the dataset should be told in the paper. This can be used to determine whether the classification task has a long tail.
6. We hope that the author share the dataset used in the manuscript to peers.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Please find in the attachment my comments, considerations, and questions.
Comments for author File: Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The paper presents a new method based on the scattering-point-guided RPN and RoI contrastive loss for ship detection in VHR SAR satellite images.
The research has been widely validated with the state of art and the experimentation is complete. Despite this, the paper is not well organised and it is difficult to be followed. I have some comments to be done before the acceptance.
Major comments.
I would considerably reduce the introduction and the state of art by summarizing the most important concepts, especially the ones that help to understand the studied research. Presented in this way, it seems completely disconnected from the rest of the paper.
For a clearer presentation, I would put from section 4.1 to section 4.3 in the methodology. In particular, 3.1 for me should be the dataset, then the methodology, the implementation details and the validation. Then, section 4 would consist in the results. Moreover, lines 437 to 442 should be put in the new section 3, at the end. They are not results, but part of the entire methodology.
The discussion and the conclusion is really poor, in my opinion. You could merge the two sections in a single one called "Discussion" but I propose the following alternative. In the result section, you should present what you obtained with the proposed method compared to the exsisting ones objectively. Then, your considerations about the results (like 442-443 lines "surpasses all comparison methods, which is very competitive compared to the current state-of-the-art methods" or from line 451 "our 450 method shows immense superiority and significantly reduces false alarms and missing 451 ships....", just to mention a few) should be in the discussion.
Minor comments.
References should be cited as [1,2] or, if they are more than 2, as [4-9]. Please check all the manuscript.
Line 70: You state "Ships in SAR images often have arbitrary orientations in both the near-shore and off-shore scenarios". This is true also for optical or other kind of images. I would say that due to the SAR geometry acquisition, there could be different orientations.
Line 88: I think there is a typing error. Do you mean show characteristics or highlight characteristics?
Line 93: "Ship target" the s should be lowercase.
Line 96: when you say "the SAR imaging conditions change" in my opinion, you should better clarify. What are the factors which influence the conditions? The backscattering could be different also in two consecutive acquisitions because of the speckle, the response of the objects, the wavelength of the incidence signal, the polarization, the shape, the roughness, the water content? What are you referring to?
Line 100: to confront or was it to overcome, to solve...?
Line 121: put ":" and not ".". Then start the itemize with lowercase letters and separate each item with ";".
Lines 223-224: and the second stage? Please explain.
Line 225: Figure2 -> put the space between figure and 2.
Figure 2: you should clarify the starting point in the figure.
Line 271: Maybe for those not familiar with the subject, explaining why the response value greater than 0.01 times the maximum one is considered.
Line 289: too many "offset". Please write it better.
Line 293: the comma should stay after the final "p" of the equation.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
