LMSD-YOLO: A Lightweight YOLO Algorithm for Multi-Scale SAR Ship Detection

Round 1

Reviewer 1 Report

1.     Introduction: The author should state why you start the lightweight YOLO work as some previous works have finished. Or, what prompted you to start the study? The authors should give a more concise reason.

2.     Line 150-151, the author needs to explain exactly why these modules can reduce the number of parameters in the model;

3.     Line 159-161, why the DSC module is used to reduce the overall computational load of the module? The DSC model is popular in the deep learning. The author often gives us the results, while we want to know the reason.

4.     Why the ACON family functions can effectively prevent neuron death in the process of large gradient propagation?

5.     Line 226-228, how could you know it won’t increase the time consumption? Not clear.

6.     Line 231, how do you know it can make easier for the model to converge?

7.     The training set and testing set of the three datasets are randomly divided with a ratio 8:2, whether is the ratio higher?

8.     4.1. Effect of DBA module: The experimental results only give graphs of the convergence process of train and do not show the indicators obtained from test, and there is no necessary connection between the indicators of training and the performance of the model. At the same time, it can be seen in the picture that the mAP of each model is similar, whether it can be said that the DBA module has little effect on the improvement of the model.

9.     Table 7, line number is putting the wrong location.

10.  In the conclusion part of the experiment, for example, in Figure 11 (e4) there is a missed detection, however, the paper describes that there is no missed detection (small target) in Line 574-575, which is not rigorous enough.

11.  Tab.8, six models not five models.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Dear Author:

I appreciate your proposal to evaluate this article and your interest in my participation.

 

The structure of the article seems correct to me and is: 1. Introduction, 2. Methodology 3. Experiments 4. Results and 5. Conclusions.

1. Introduction

The authors have done a good job of reviewing the state of the art in the literature referenced in References. The introduction seems to me to be well developed and substantiated with the bibliographical references being very up-to-date.

The references are correct and complete, being varied and updated.

2.Methodology

The methodology is well explained and the mathematical foundations are well exposed.

Tables and figures are correct.

3.Experiments

The development of my experiments through a training platform and a testing platform seems very well developed to me.

Tables and figures are correct.

4.Results

The results obtained are in agreement with the applied methodology. All this seems correct to me.

I find the results very interesting.

Tables and figures are correct.

5. Conclusion

The conclusions are consistent with the proposed objectives.

 

I recommend the publication of this manuscript.

 

Best regards

Author Response

Dear Editors and Reviewers,

      We sincerely thank the editors and reviewers for the opportunity to submit the revised version of our manuscript entitled “LMSD-YOLO: A lightweight YOLO algorithm for muti-scale SAR Ship detection”. All the comments and suggestions are valuable and helpful for revising the manuscript and guide our research. We have carefully studied the reviewers’ comments and made corresponding explanations and adjustments. Note that our responses are given directly afterward in red text and the changes to the manuscript are also colored in red text. Our responses are organized and presented item by item. We hope that the responses have covered all the points raised by the reviewers, and the revised version is suitable for publication.

 

Best Regards,

Yue Guo

School of Electronic Science and Engineering, National University of Defense Technology, Changsha, 410073, Hunan, P. R. China.

Tel: +86 159-1259-7943

E-Mail: [email protected]

Author Response File: Author Response.pdf

Reviewer 3 Report

The aim of this paper is quite interesting. Study results are very convincing and may add to the existing knowledge. However, the following minor flaws need to be taken care of prior to publication acceptance:

1.          Several keywords duplicate the same as in the paper title. Other selections should be re-chosen.

2.          All acronym names should be defined when first appear and be used thereafter.

3.          In Line 113, the word “contributions” may be changed to “objectives”.

4.          The last paragraph of the introduction section is redundant.

5.          In Line 592, the phrase “large raw large image data” is a bit unclear.

6.          References No. 1 and 19 lack year specification.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

A lightweight LMSD-YOLO model is proposed to solve the problem of SAR ship target recognition in coastal and densely distributed areas. The main contribution of this paper is to complete the hardware deployment and testing by optimizing the network structure and the design of basic modules in advance to ensure the accuracy. The experimental results show that the proposed method is effective and available.

I recommend that the paper be published after some minor modifications.

First of all, in the introduction, when the author reviews various SAR ship target recognition methods, some recent interesting research literature has been ignored, including, for example, “Spatial singularity expert domain multiresolution Imaging Based SAR ship target detection method" in IEEE Trans.  Geosci.  Remote.  Sens. 60: 1-12 (2022), doi: 10.1109/TGRS.2021.3113919. and（2）BiFA-YOLO: A Novel YOLO-Based Method for Arbitrary-Oriented Ship Detection in High-Resolution SAR Images, in Remote Sensing.

Secondly, the author considers the rotation loss in the loss function, but the experimental part does not give the detection results of the target direction. Can the network proposed in this paper also be used for SAR ship target orientation detection?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf