Next Article in Journal
Unification of a Global Height System at the Centimeter-Level Using Precise Clock Frequency Signal Links
Next Article in Special Issue
Multi-Oriented Enhancement Branch and Context-Aware Module for Few-Shot Oriented Object Detection in Remote Sensing Images
Previous Article in Journal
Evaluating the Effect of Training Data Size and Composition on the Accuracy of Smallholder Irrigated Agriculture Mapping in Mozambique Using Remote Sensing and Machine Learning Algorithms
Previous Article in Special Issue
A Class-Incremental Learning Method for SAR Images Based on Self-Sustainment Guidance Representation
 
 
Article
Peer-Review Record

ESarDet: An Efficient SAR Ship Detection Method Based on Context Information and Large Effective Receptive Field

Remote Sens. 2023, 15(12), 3018; https://doi.org/10.3390/rs15123018
by Yimin Zhang †, Chuxuan Chen †, Ronglin Hu * and Yongtao Yu
Reviewer 1: Anonymous
Reviewer 2:
Remote Sens. 2023, 15(12), 3018; https://doi.org/10.3390/rs15123018
Submission received: 4 May 2023 / Revised: 28 May 2023 / Accepted: 5 June 2023 / Published: 9 June 2023

Round 1

Reviewer 1 Report (Previous Reviewer 4)

The work presented An Efficient SAR Ship Detection Method Based on Context Information and Large Effective Receptive Field. Overall, the structure of this paper is well organized, and the presentation is clear. The idea is interesting and the reviewer only has few comments.

1.      The reviewer is wondering whether the AI or machine learning in remote sensing methods can achieve the goal. Therefore, the reviewer suggests discussing some related works by analyzing the following papers in the revised manuscript, e.g., 10.1109/TGRS.2020.3015157, 10.1109/TGRS.2020.3016820。

2.      Please clarify the contributions, why this method is important?

3.     Some English writing typos should be corrected.

4.     Some SOTA detection methods in remote sensing should be discussed or compared, e.g., ORSIm detector, UIU-Net.

5.     It is well-known that the data usually tend to suffer from various degradation, noise effects, or variabilities in the process of imaging. Please give the discussion and analysis by referring to the paper titled by e.g., An Augmented Linear Mixing Model to Address Spectral Variability for hyperspectral unmixing. The reviewer is wondering what will happen if the proposed method meets the various variabilities.

6.     Some future directions should be pointed out in the conclusion.

The work presented An Efficient SAR Ship Detection Method Based on Context Information and Large Effective Receptive Field. Overall, the structure of this paper is well organized, and the presentation is clear. The idea is interesting and the reviewer only has few comments.

1.      The reviewer is wondering whether the AI or machine learning in remote sensing methods can achieve the goal. Therefore, the reviewer suggests discussing some related works by analyzing the following papers in the revised manuscript, e.g., 10.1109/TGRS.2020.3015157, 10.1109/TGRS.2020.3016820。

2.      Please clarify the contributions, why this method is important?

3.     Some English writing typos should be corrected.

4.     Some SOTA detection methods in remote sensing should be discussed or compared, e.g., ORSIm detector, UIU-Net.

5.     It is well-known that the data usually tend to suffer from various degradation, noise effects, or variabilities in the process of imaging. Please give the discussion and analysis by referring to the paper titled by e.g., An Augmented Linear Mixing Model to Address Spectral Variability for hyperspectral unmixing. The reviewer is wondering what will happen if the proposed method meets the various variabilities.

6.     Some future directions should be pointed out in the conclusion.

Author Response

Thank you for your comments. We have made point-to-point revisions to the article in response to these comments. The details of the revisions are shown in the revision reports.

Author Response File: Author Response.pdf

Reviewer 2 Report (Previous Reviewer 3)

This paper develops an efficient SAR ship detection method based on context information and a large effective receptive field (ERF) based on deep learning architecture. The main goals and topics of the paper are interesting, important, and up-to-date. The abstract is well structured and the keyword selection is appropriate. The overall study also seems to have been done in a good enough way from a methodological point of view. However, the authors are encouraged to study the following comments and suggestions to revise the manuscript accordingly. Specific comments are as follows.

1.The beginning of the introduction is not good, and it should be reconstructed. Related work is necessary to increase the analysis of the literature published in the last two years and compare the differences between these works and the work in this paper. 

2.Lines 24-30, lines 42-45 & lines 168-174 need to add references.

3.Clearly explain the research problems or the challenges for ship detection. Line “69” to line “95” the objective of this paper is descriptive. Make it more logically simple and specific for the reader.

4.The methodology should be explained in a more analytical way to strengthen this paper. In this section, most of the figures need some more logical explanation for readers, so I suggest adding more explanation to clarify the applicability of the proposed model.

5.As the authors have compared with the SOTA detectors so far tables 6, 7, and 8 need some logical explanation for the reader. The comparative experiment in this article only compares the one-stage detection model without comparing the two-stage detection model, and additional experiments are needed. Further explanation is needed to determine whether it can significantly improve the problem of large computation for high-resolution images compared with a SOTA detector.

6.It would be helpful to include the estimated runtime for specific datasets using ESarDet in this configuration. Regarding the improvement of the loss function and the role of CIoU, would it be better to reflect some of them in Ablation Studies as well?

7.How fast is the processing speed of the algorithm proposed in this paper? It is recommended to analyze the time complexity of the algorithm.

8.The results and discussion sections should be updated to reflect the study's derivations. In addition to the problems outlined above, the results of the experiment section have a weak hierarchical organization and are less understandable. The authors should describe it more logically. The equations in section 4.4 are not explained in detail, and all the equations need to properly cite in the main text as well.

9.The conclusion should be re-written . 1) explicitly describe the essential features/advantages of the paper, and 2) describe the limitation(s) of the paper.

This paper develops an efficient SAR ship detection method based on context information and a large effective receptive field (ERF) based on deep learning architecture. The main goals and topics of the paper are interesting, important, and up-to-date. The abstract is well structured and the keyword selection is appropriate. The overall study also seems to have been done in a good enough way from a methodological point of view. However, the authors are encouraged to study the following comments and suggestions to revise the manuscript accordingly. Specific comments are as follows.

1.The beginning of the introduction is not good, and it should be reconstructed. Related work is necessary to increase the analysis of the literature published in the last two years and compare the differences between these works and the work in this paper. 

2.Lines 24-30, lines 42-45 & lines 168-174 need to add references.

3.Clearly explain the research problems or the challenges for ship detection. Line “69” to line “95” the objective of this paper is descriptive. Make it more logically simple and specific for the reader.

4.The methodology should be explained in a more analytical way to strengthen this paper. In this section, most of the figures need some more logical explanation for readers, so I suggest adding more explanation to clarify the applicability of the proposed model.

5.As the authors have compared with the SOTA detectors so far tables 6, 7, and 8 need some logical explanation for the reader. The comparative experiment in this article only compares the one-stage detection model without comparing the two-stage detection model, and additional experiments are needed. Further explanation is needed to determine whether it can significantly improve the problem of large computation for high-resolution images compared with a SOTA detector.

6.It would be helpful to include the estimated runtime for specific datasets using ESarDet in this configuration. Regarding the improvement of the loss function and the role of CIoU, would it be better to reflect some of them in Ablation Studies as well?

7.How fast is the processing speed of the algorithm proposed in this paper? It is recommended to analyze the time complexity of the algorithm.

8.The results and discussion sections should be updated to reflect the study's derivations. In addition to the problems outlined above, the results of the experiment section have a weak hierarchical organization and are less understandable. The authors should describe it more logically. The equations in section 4.4 are not explained in detail, and all the equations need to properly cite in the main text as well.

9.The conclusion should be re-written . 1) explicitly describe the essential features/advantages of the paper, and 2) describe the limitation(s) of the paper.

Author Response

Thank you for your comments. We have made point-to-point revisions to the article in response to these comments. The details of the revisions are shown in the revision reports.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report (Previous Reviewer 4)

The authors have well addressed the reviewer's concerns. No more comments.

Reviewer 2 Report (Previous Reviewer 3)

The paper has revised all the issues that I proposed, and I think it can be published in this version.

The paper has revised all the issues that I proposed, and I think it can be published in this version.

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

This paper proposed a new SAR Ship Detection method based on YOLOX-tiny. Several improvements have been made, including the proposed CAA-Net, A2SPPF and A2CSPlayer. All these strategies are designed for exploring contextual information or enhance the fusion of feature maps from various scales. Although very high average precision is achieved on three public SAR ship datasets, DSSDD, SSDD, and HRSID. Several major problems need to be revised:

1. the novelty of the paper is not enough. The idea of the proposed CAA-Net, A2SPPF and A2CSPlayer are very common in the field of deep learning.

2. The entire article does not analyze the characteristics of the ships in the SAR imagery. The authors mentioned several challenges faced in SAR ship detection, including complex backgrounds, large scale variations, small scale targets, and other challenges. Then, how this article addresses these challenges.

3. The biggest problem is the experimental analysis part. The analysis is too simple, especially the analysis of Ablation experiment, and I can't see some valuable conclusions.

 

4. All the methods chosen for comparison come from computer vision and some SOTA models in the field of SAR ship detection should be included for comparison.

Reviewer 2 Report

You should check language and rephrase some parts. For example, in the Abstract you have two subsequent sentences beginning with "On the one hand, we". Or line 78:  To increase xxx and avoid xxx, ....???It cannot be end of the sentence.

 

F1 − score in eq (21) looks like subtraction (math operation), and not like variable name, eg. F1_score or F1score or similar. 

 

You should provide some statistics to prove significance of your findings.

Reviewer 3 Report

This paper seems quite sound and clearly described. The authors propose an ESarDet, an efficient SAR ship detection method based on context information and a large effective receptive field. Compared with existing detection methods, the model presented in this paper has more accurate and efficient. This work is valuable for target identification in RS imagery. Experiments on these datasets are provided to demonstrate that the method is effective and leads to improved performance. However, there are still some questions that need to be issued before going for publication. The comments and suggestions are;

1.The introduction part needs to be expanded by including the most recent information and studies on why this study is necessary and what recent research on target detection using Deep Learning has discovered.

2.Clearly explain the research problem or the challenges for ship detection the paper is addressing.

3.Hyperparameters have a significant impact on machine learning methods. The authors should add a new part in the paper to discuss how to choose the hyperparameters for their method or other deep learning models.

4.Lightweight networks are an effective way to solve the problem of detection on platforms with limited computing resources, but the detection speed of the model should also be considered. In terms of evaluation indicators, YOLOX-tiny is a lightweight network, and FPS should be used as an indicator to evaluate the performance of the model.

5.The figures should be clear and self-explanatory, and the authors should check the whole manuscript. Before use, all variables in the method and equations should be clearly structured and clearly described.

6.Presentation of the proposed method is generally poor, even though some parts of the presentation contain enough details and are clear and some are not such as figures 2, 3, 5, 6, and 8. ? A clear and logical description is needed.

7.How are the parameters of the model set? Is there any basis or reference?

8.The equations are not explained in detail; a detailed explanation is needed of what each part of the formula represents.

9.The literature review looks faded, and it should be expanded by adding new techniques for similar types of application analysis on a global scale to make this research more relevant. There are several recent works that benefit from efficient deep learning to address ship detection such as https://www.frontiersin.org/articles/10.3389/fmars.2022.1086140/full. For a detailed discussion of the use of deep learning in ship detection, this work should be addressed in the related work section.

10.Line “166” it is suggested to insert text first instead of the figure for a better look, furthermore figure 1 needs some more logical explanation to support the suggested method used in this paper. 

11.The results and discussions section should be revised to demonstrate the derivations of the current research performed. The validity of the presentation and testing results in the experimental part of the paper is not verified. In the part on evaluation indexes and visualization of testing results, it is recommended to list the utilized evaluation indexes as the criteria for umpiring the outcome of the following experiments and to utilize graphs to redirect the validity of the suggested approach.

12.Line “349” confirmed this ⌊.⌋ whatever it is correct or not because it does not appear in the equation?

13.The reference format needs to be modified, some reference’s year is bold while some are not, it is recommended to check against the journal template, and a uniform template is required for all references. 

14.Some occasional English language and style errors can be found. 

Reviewer 4 Report

This paper proposed An Efficient SAR Ship Detection Method Based on Context Information and Large Effective Receptive Field. Overall, the structure of this paper is well organized, and the presentation is relatively clear. The idea is interesting and potential. However, there are still some crucial problems that need to be carefully addressed before a possible publication. More specifically,

1.       The motivations or remaining challenges are not so clear or what kinds of issues or difficulties are this task that is facing. Please give more details and discussion about the key problems solved in this paper, which is largely different from existing works.

2.       A deep literature review should be given, particularly some currently advanced deep learning models or AI models in remote sensing. Therefore, the reviewer strongly suggests discussing some related works by analyzing the following papers in the revised manuscript, e.g., 10.1109/TGRS.2020.3016820, 10.1109/TGRS.2020.3015157.

3.       Please clarify the contributions to this field, for example, which are the existing ones and which are your own ones? The elements used in this work seem to be existing methods. What are your newly-added values.

4.       How about its computational complexity compared to existing methods.

5.       Some SOTA detection methods should be mentioned further, e.g., ORSIm detector, UIU-Net.

 

6.       Some future directions should be pointed out in the conclusion.

Back to TopTop