CME-YOLOv5: An Efficient Object Detection Network for Densely Spaced Fish and Small Targets

Round 1

Reviewer 1 Report

This manuscript proposed an approach to improve the efficiency of fish  population detection, especially for the high-density, mutual blocking and the small target groups. The experimental results showed that the mAP_0.5 of the proposed algorithm reached 94.9%, which is 4.4 percentage points higher than that of the YOLOv5 algorithm, and the number of fish and small target detection performance was 24.6% higher. The optimization method may serve as an alternative or supplemental method for fishery resource investigation. I recommend minor revision of this manuscript. I hope my suggestions will help the authors to improve the manuscript and make it more suitable for publication.

1. It is suggested to further increase the application scenarios of this method in the future to solve the problems of dense and small target recognition. More contents could be added in the Discussion section.

2. Conclusion section, I feel that some sentences of this section have been repeated from those in the ‘Result’ and ‘Abstract’.

3. The width of tables 1, 5 and 6 is inconsistent, and the width of tables 2, 3, 4 and 7 is inconsistent. It is recommended to unify the width and check the full text

4. The title serial number is reused. Check it.

5. Check the format of the references, e.g., pages, capital letter, and volume.

Author Response

Dear reviewer,

      We thank you very much for your careful and thorough reading of this manuscript, as well as your thoughtful comments and constructive suggestions. The reply letter has been uploaded in the form of a file, please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The manuscript proposes an improved yolov5 algorithm for the detection of densely populated fish from the underwater images. The topic is very interesting and in the scope of Water. Authors proposed a novel method and achieved the higher detection accuracies than the literature. The manuscript is very well written and presents significant scientific contents in terms of innovative development within the computer vision. Followings are few of my concerns which may improve the overall structure and readability of the manuscript.

 

1.     The title of the manuscript may need a revision. Rather than saying “Improved YOLOV4”, I would suggest authors to formulate a proper name for their algorithm and always use that abbreviation. For example, “DenseObjDetNet” or something. 

2.     Abstract lacks the motivation information; I would suggest authors to add couple of lines related to motivation of performing such study. Why underwater fish detection is important? Why existing methods are not up to mark? And then start with the proposed algorithm.

3.     The keywords should reflect the Abstract and should only contain the words which are used within the Abstract. Please cross-check if any such keyword exists which does not appear in Abstract, either remove that or adjust in the Abstract.

4.     Do not use Abbreviations in the Abstract, always use the full form XXX (XXX).

5.     Introduction section is not written comprehensively and needs a bit of effort to improve. Introduction should follow the following structure: 

a.     What is the significance of the underwater fish detection? Provide some statistics to validate the importance.

b.     A very brief summary of one paragraph about what has been done generally in terms of underwater fish detections. What different approaches are being used in practice and what are the limitation of those approaches?

c.     A very brief introduction of the computer vision technology in context of underwater fish detection, specifically focussing on the small target detection. Finish with the paragraph with the limitations of the existing computer vision models.

d.     A paragraph of the proposed architecture and how it is different from the existing methods. Highlight or preferably list the anticipated contributions of your research.

e.     Finally, provide the layout of the manuscript in the ending paragraph of the introduction.

6.     A detailed literature review section with critical insights into the existing methods is clearly missing. A brief literature review contents can be found within the introduction, however needs to move to a separate “Literature Review” section. Authors may consider sub-sections within the literature review section (i.e., advancements in fish detection methods and state of the art in the small targets object detection methods) where they can discuss the literature in detail. I would suggest at least a one and half page section with some subjective criticism on the literature to provide readers with an insight into the state of the research within this domain of underwater fish detection. 

7.     Authors may consider merging existing section 2 and section 3 with “The Proposed CME-YOLOV5”. Consider the very first comment, if authors come up with a name, this section should be titled on the formulated name like “DenseObjDetNet – An Improved Yolov5 Model” so something similar.

8.     It is recommended to keep the results and discussions as separate section from the experimental protocols. I would suggest the following structure to improve the readability:

                                               i.     Introduction

                                             ii.     Literature Review

                                            iii.     DenseObjDetNet – An Improved YOLOV5

                                            iv.     Dataset

                                              v.     Experimental Protocols and Evaluation Measures

                                            vi.     Results and Discussions

                                           vii.     Conclusion

9.     The correct notation is: [email protected] and not the mAP_0.50. Please correct throughout the manuscript.

10.  Figure 3, authors may provide the improved resolution graphics, preferably the vector graphics.

11.  Table 5, authors need to provide the notes below the table about each model they are using Model 1 till Mode 6.

12.  Conclusions should talk a bit about the limitations of the proposed model and potential future research directions. 

Author Response

Dear reviewer,

      We thank you very much for your careful and thorough reading of this manuscript, as well as your thoughtful comments and constructive suggestions. The reply letter has been uploaded in the form of a file, please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Most of the comments provided have been addressed.