Offshore Oil Platform Detection in Polarimetric SAR Images Using Level Set Segmentation of Limited Initial Region and Convolutional Neural Network

Round 1

Reviewer 1 Report

I think the innovation of this paper is not very prominent. However, the biggest problem is that the author gives the comparison results of each detail in the experimental part, and does not give the comparison results of the whole process, which cannot reflect the advantages of the entire algorithm process. In addition, Figure 7 shows the results of the proposed method in different scenarios, but this figure does not explain any problems. Please carefully reorganize the entire experimental section.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Original Submission

Recommendation

Minor revision

Comments to Author:

 

Title:  

 

 

Offshore Oil Platform Detection in Polarimetric SAR Images Using Level Set Segmentation of Limited Initial Region and Convolutional Neural Network

 

Overview and general recommendation.

 

This paper is demonstrating usage of SAR data and some important applications. This technique is very important and somehow accurate and has lots of important applications. First of all, as a person with more than 20 years familiarity with SAR/RADAR data, I like this paper very much; but as a scientist, I have to say the truth about the material and to be honest. Very good materials have been explained; I really like this paper.

The Abstract is OK: summarizing the idea and concepts inside the paper; English is fair; but, I think it is better to give it to a native person to review (it is a big must).

Introduction is fair. I think in some positions, some important corrections must be done; leak of some Refs... please fix them! Pls improve the quality of some Figs => they are very bad; unacceptable (I do not accept this!).  

I have some questions too: 1. how the polarizations (HH, VV, HV, or VH) differ on this project? 2. How the high resolution data such as CSK or TSX act on the data? 3. I think for backscattered methods, it is better to use many and compare them; what do you think? 4. How did you handled speckle problems? 5. Can we just use ordinary image processing (for instance what we do in python) to detect the objects and contour them? Will you pls explain a bit?

I like this paper very much: good experiments have been done; however, I think this work must be improved and lots of things to do (mainly on backscattered methods); agreed? I think you have to be very careful on that. Pls do these primary corrections that I have said, and then I will go over the paper again…

 

Detailed comments:

Line 3. Write PolSAR. And in line 23, and 109 too.

Line 11-14. … other traditional methods??? What do you mean? I think it is better explain more on this.

Line 27. I have a question: can we use PSs/SBAS to get high backscattered signals from platforms?  As this line and few lines earlier say.

Line 122. Rephrase pls (mainly?).

Line 159. What if not complex Wishart distribution conditions meet? Why not use two scale methods?

Table 2. It is not  “angle of incident” but it is incidence angle. AOI stands for area of interest. BUT as you wish… Also incidence angle changes for entire image what is the number you have given? Also what is the size dimensions: is it pixels? Meters? KM?

Fig11… Quality? All of them…

 

 

 

 

 

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper proposes a method for detection of oil Offshore Oil Platform using polarimetric SAR images. If I understood correctly the method is composed of object detection method (GOPCE), segmentation method, the smallest enclosing circle (SMSEC) and classification method that separates ships and oil platforms.

1. Is the position of oil platforms known in advance? What is the purpose of the algorithm? Who is the end user? The motivation is missing.
2. The polarimetric data can well distinguish between the natural object, see clutter and human made objects. The classification methods using general convolutional neural networks can classify objects, can detect objects and can be used for object recognition. Off course, question is why you are not using object recognition method, like Yolo or R-CNN to detect object of interest.
3. The oil platform is detected by using the segmentation. How the (4) is solved?
4. Why are you using the GOPCE-Based approach to ship detection presented in [13]. Is this step necessary? What are the modifications compared to the original algorithm [13]?
5. It is not clear why you need to group all segmented pixels using the circle based enclosing algorithm? This could be integrated in to GOPCE-based method.
6. The training data and information how the classification-based method was trained is missing.
7. The objects for detection are not presented at the beginning of the paper (SAR image with targets). The object of interest should be visualized and discussed, why the ship detection method can not recognize the oil platform and what are statistical properties of ships and oil platformy when using the PolSAR data. The ship detection methods should not be able to detect oil platforms. This could improve understanding of the problem and better justification of used methods.
8. All tables are not clearly present the experimental results. In Table 5 caption is missing, what table represents.
9. A set of 19 targets is very small training set to verify the efficiency of the proposed method.
10. Why the CNN classifier is compared to the SVM based classifier? If this section remains than you should do very similar to all other used methods.
11. Comparison with the object based CNN methods is missing.
12. Images are very small. Have a feeling that the object consists vary small number of pixels.
13. Literature review should be updated with the recent the state of the art.

 

The proposed method is a combination of existing methods, therefore the novelty is very marginal. Authors should improve the paper, especially in the section that will describe the data presentation and interpretation, therefore, major revision is needed.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

No more comments.

Author Response

Dear Reviewer 1:

Thank you for your valuable comments on our manuscript. After carefully studying your comments, we have carefully revised the manuscript.

 

[Reviewer 1 Comment 1]:

No more comments.

[Response to Comment 1]:

Thanks very much for your valuable comments.

 

Yours sincerely,

Chun Liu

March 31, 2022

Reviewer 3 Report

Dear Authors,

all comments are not answered within the revised manuscript.

Please, in the conclusion section write that only 10 polarimetric datasets were used and that the experimental results could be improved.  

After minor revision, paper can be accepted.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf