Next Article in Journal
Airborne SAR Imaging Algorithm for Ocean Waves Based on Optimum Focus Setting
Next Article in Special Issue
Sea Surface Roughness Observed by High Resolution Radar
Previous Article in Journal
A Comparison of Three Sediment Acoustic Models Using Bayesian Inversion and Model Selection Techniques
Previous Article in Special Issue
A Wind Speed Retrieval Model for Sentinel-1A EW Mode Cross-Polarization Images
 
 
Article
Peer-Review Record

Extension of Ship Wake Detectability Model for Non-Linear Influences of Parameters Using Satellite Based X-Band Synthetic Aperture Radar

Remote Sens. 2019, 11(5), 563; https://doi.org/10.3390/rs11050563
by Björn Tings *, Andrey Pleskachevsky, Domenico Velotto and Sven Jacobsen
Reviewer 1: Anonymous
Reviewer 2: Anonymous
Reviewer 3: Anonymous
Reviewer 4: Anonymous
Remote Sens. 2019, 11(5), 563; https://doi.org/10.3390/rs11050563
Submission received: 29 January 2019 / Revised: 24 February 2019 / Accepted: 25 February 2019 / Published: 7 March 2019
(This article belongs to the Special Issue Sea Surface Roughness Observed by High Resolution Radar)

Round 1

Reviewer 1 Report

Dear Authors,

thank you for your work but the reviewer thinks the paper should improve descriptions. 

Why did not you face the azimuth shift problem? This question is till open.

You wrote a code, and you tested it, the results of the experiments should cover your results (as you have already done) and provide a measure of the complexity of the algorithm you wrote and tested.


Author Response

Reviewer 1

Comments and Suggestions for Authors

Dear Authors,

thank you for your work but the reviewer thinks the paper should improve descriptions.

Authors’ answer: We thank you for your time and for the suggestions provided in order to improve the descriptions of our work. Your thoughts are taken into account in the revised version of the manuscript.

Why did not you face the azimuth shift problem? This question is till open.

Authors’ answer: In Section 2 “Materials and Methods” the flow-chart in Figure 1 describe each single step performed to obtain the results of this study. Some steps involve procedures or algorithms already published by us and for those the detailed description is left out to leave space to the current objectives and analysis and the references are given. In particular, the azimuth shift problem is related to the association of a ship in SAR image to the AIS message. In the current version of the manuscript the wake labelling procedure is describe as “…an automatic intersection of AIS with the SAR images was executed to assign AIS messages to image regions. A manual correction of these colocations was performed to let the unreliable AIS data fulfill ground truth requirements, which means colocations have been discarded in case of large amounts of artefacts like ambiguities or marine objects being present. Then on the basis of these two datasets co-located in space and time a manual search for moving vessels was conducted (Figure 1:A.1). During the search the background of the moving vessels was checked for unambiguous visibility or non-visibility of wake signatures. By doing so to each wake sample either the class label “detected” or “not detected” was assigned (Figure 1:A.2). Detailed information about the manual inspection procedure can be found in [14].” The wake signatures are not affected by azimuth shift and correspond to the AIS positions. So, in our opinion there are no reasons to face the azimuth shift because is not really a problem.

You wrote a code, and you tested it, the results of the experiments should cover your results (as you have already done) and provide a measure of the complexity of the algorithm you wrote and tested.

Authors’ answer: The complexity of the procedure applied to retrieve the scientific results of this paper is well described by the flow-chart in Figure 1. The extraction of all influencing parameters involves the utilization of wind and wave algorithms developed in house for TerraSAR-X images and previously published. In our opinion is out of the scope of this paper to provide the computational burden of those algorithms. The objective of this paper is to quantify numerically the impact of the influencing parameters on the wake detectability in TerraSAR-X images and not to have an operational wake detection algorithm. Therefore, even if the process of collecting the training parameters is long and computationally expensive, it doesn’t really influence the objectives posed. Training the SVM classifier it is done in few seconds as well as using it for the applications like minimum vessel velocity.


Reviewer 2 Report

Good work. Accepted.

Author Response

Reviewer 2

Comments and Suggestions for Authors

Good work. Accepted.

Authors’ answer: Thank you for your positive comment.


Reviewer 3 Report

Lines 229-230 “It turned out that only a polynomial kernel with a degree of two can outperform the linear model”. It is not clear which range of polynomial degrees were tested in order to conclude that the best was the one with polynomial degree 2.

Line 238 “Gamma-parameters…” should be expressed in another way e.g. “Gamma-parameter values”. The expression “Gamma-parameters ranging from 0.001 and 0.1 were tried and best performance is achieved…” should be expressed in another way in order to be right grammatically.

Line 234 Different cost-parameters ranging from 0.01 to 100” (10000 combinations),

Line 238 “Gamma-parameters ranging from 0.001 and 0.1” (100 combinations)

Line 240 “coef0-parameter was performed over the value range of 0 to 1000” (1000 combinations)

So, 10000 X 100 X 1000 = 1 billion combinations were tested? How this is possible?


Author Response

Reviewer 3

Comments and Suggestions for Authors

Lines 229-230 “It turned out that only a polynomial kernel with a degree of two can outperform the linear model”. It is not clear which range of polynomial degrees were tested in order to conclude that the best was the one with polynomial degree 2.

Authors’ answer: The sentence is edited to provide the actual meaning. The word “only” might have created such confusion. Here we mean that using a polynomial kernel with degree of two outperform our previously used linear model (e.g. more flexibility and nonlinear dependency to the influencing parameters). We do also provide motivations for not going with higher polynomial degree or complexity. Lines 230-233 in the text “First, a much higher complexity, induced by higher degree of the polynomial kernel or due to radial-basis or sigmoid-kernel, leads to overfitting [36]. Second, for all the nine selected influencing parameters only one detectability peak is expected, therefore a polynomial kernel with a degree of two is sufficient to model this one peaked maximum.

Line 238 “Gamma-parameters…” should be expressed in another way e.g. “Gamma-parameter values”. The expression “Gamma-parameters ranging from 0.001 and 0.1 were tried and best performance is achieved…” should be expressed in another way in order to be right grammatically.

Authors’ answer: Thank you for your suggestions. The text has been changed accordingly.

Line 234 Different cost-parameters ranging from 0.01 to 100” (10000 combinations),

Line 238 “Gamma-parameters ranging from 0.001 and 0.1” (100 combinations)

Line 240 “coef0-parameter was performed over the value range of 0 to 1000” (1000 combinations)

So, 10000 X 100 X 1000 = 1 billion combinations were tested? How this is possible?

Authors’ answer: Thank you for pointing out this material mistake. The step applied in each range of the hyperparameters’ values was missing in the text and it is now explicitly mentioned.


Reviewer 4 Report

The name of the 9th parameter listed in Table I is not visible in the pdf file. Please check. 

Author Response

Reviewer 4

Comments and Suggestions for Authors

The name of the 9th parameter listed in Table I is not visible in the pdf file. Please check.

Authors’ answer: Thank you for pointing out this material mistake. The missing name of the 9th parameter is now included in Table I.


This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.


Round 1

Reviewer 1 Report

Dear Authors,
the reviewer thinks the paper should improve some descriptions. A) Your strategy to face the azimuth shift problem is not clear. Please, provide a detailed description where you explain how you resolve this problem. B) You should present the computational time (in relation to the feature of the input image) and the complexity of your algorithm with your results.

Reviewer 2 Report

Dear Authors,

This paper  is technically correct,

The experimental results must be better explained and showed.

A small point to point revision chart has been provided.

Good work

Comments for author File: Comments.pdf

Reviewer 3 Report

The paper is interesting but there are issues that must be addressed.

-The phrase in lines 105-107: “In fact it turned out that only a polynomial kernel with a degree of two can outperform the linear model. The reason is that a higher complexity, induced either due to higher degree of the polynomial kernel or due to radial-basis- or sigmoid-kernel, leads to overfitting” should be reconsidered by the authors. It seems as an arbitrary explanation without mentioning any references.

-Furthermore, the authors have not presented the experimental results of testing different SVM Kernels and the range of the parameters (C, gamma, etc.) that have been tested (E.g. gamma from 0.1 to 1?).

-The lines 103-111 should be included in the results section and not in the “Materials and Methods” section.

-Also, the calculation of the absolute correlation coefficients presented in figure 1, table 1, should be in the results section.  

-In lines 97-98 the authors refer “The same approach is adopted for this study, but a Support Vector Machine (SVM) classifier is trained on all influencing parameters together”. It is not clear why the authors have previously calculated the correlation coefficients of the influencing parameters since they have used all the influencing parameters together as input factors in the SVM models.  

-In line 15 the authors refer “a large dataset of manually identified wake samples”. More information about how the dataset was created. Which was the time period of the observations? What kind of ships were selected to be observed and why? etc.

-The authors should add more theoretical information about the SVMs.

- It is not explained enough why the article makes an original contribution to new knowledge. The results are not discussed enough. The authors should compare their methodology with similar studies and explain more why their study contribute to new knowledge.


Reviewer 4 Report

The paper shows an interesting analysis of the effect of some parameters (SAR-related, Ship route-related and environmental-related) on the wake detectability. The results discussion starts from the study published in a previous paper [ref 10] and enrich the analysis by using a more complex model and additional parameters.

The paper deserves the publication on Remote Sensing after the implementation of some comments

 

 

Major Comments

·         In the first sections (introduction or section 2) it should be clarify the assumptions related to the wake appearance, e.g. no narrow-V wake are mentioned, when the wake is assumed detected, if only the turbulent component is imaged, the wake is classified as detected?

·         Section 2 should be re-written. Only 8 parameters are listed in Table I (on the contrary in the text it is reported that 9 parameters are listed in Table I). 12 parameters are shown in Figure 1 with the related correlation coefficient.  The readability of the section should be improved. It could be better: 1) introduce and describe the 12 influencing parameters, 2) describe the value range for each parameters, 3) describe in more details how the Figure 1 is obtained. The dataset description and the used model (SVM) could be postponed in section 3.

·         Page 4, lines 94-109: please clarify the concept on the basis of the SVM model and why it improves the results of ref. 10

·         Section 3 is quite messy. After the description of the 2D, 3D maps, it should contain a subsection for each parameter with the most significant plots. In such a way, a reader interested in a specific influencing parameters can easily reach the information

·         Section 5. The authors declare that a reversion of the method could allow an estimation of ship velocity (also in page 12, line 377). This consideration is very questionable and very interesting at same time. From Figure 9, it seems that the authors want to provide an understanding of the minimum vessel velocity as a function of incidence angle and wind, not a general rule for the ship velocity estimation. Please clarify.

 

Minor Comments

·         Introduction. The references on wake detection algorithm could be significantly improved with more recent manuscripts.

·         Page 2, lines 58: please clarify the meaning of “automatic intersection of AIS” and “manual correction”

·         Page 8, lines 230-234. It should be mention the effect of the radar observation direction

·         Page 9, lines 251-258. Please improve the English style. A lot of sentences are segmented

·         Page 9, line 264: Please clarify “this compensates to some degree”

·         Page 9, line 276-277: please clarify the English style of the sentence

·         Page 10, line 311-312: please clarify the English style of the sentence


Back to TopTop