A Real-Time Lightweight Detection Algorithm for Deck Crew and the Use of Fishing Nets Based on Improved YOLOv5s Network

Round 1

Reviewer 1 Report

The work is interesting and practical. However, it is more engineering than scientific. I doubt very much that this fits the aims and scope of Fishes magazine.

If it is still decided to print it, then please pay attention to the fact that Tables 1 and 2 are redundant; their information can be placed in the text. And also it is not clear what connection fishery has with agriculture (Lines 23-26).

Author Response

Dear Reviewer,

Thank you very much for your time involved in reviewing the manuscript and your very encouraging comments on the merits.

Q1：then please pay attention to the fact that Tables 1 and 2 are redundant; their information can be placed in the text.

Response：Thank you for your suggestion. As suggested by reviewer, we have replace Tables 1 and 2 with text.

Q2: And also it is not clear what connection fishery has with agriculture.

Response: We apologize for the language problems in the original manuscript. As suggested by reviewer, we have replaced the ‘agriculture’ with ‘Chinese economy’.

Sincerely,

The Authors

Reviewer 2 Report

A reasonable article but very long. 

Not clear why all the variations have been done. It is just a simple application of YOLOv5. To make it interesting for the readers, please make clear what you are getting from the YOLOv5 variations (off-the-shelf, without any modifications).

That way, the readers can assess how they can use the same.

Please note! You have multiple reference format errors

e.g. year???

Han, K.; Wang, Y.; Tian, Q.; Guo, J.; Xu, C.; Xu, C. Ghostnet: More features from cheap operations. In Proceedings of the

Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 1580–1589.

Need minor proofreading. 

Author Response

Dear Reviewer,

Thank you very much for your time involved in reviewing the manuscript and your very encouraging comments on the merits.

Q1：Not clear why all the variations have been done. It is just a simple application of YOLOv5. To make it interesting for the readers, please make clear what you are getting from the YOLOv5 variations (off-the-shelf, without any modifications).

Response：Thanks for your great suggestion on improving the accessibility of our manuscript. We have added a brief overview of the overall improvement program of YOLOv5s and the results that can be obtained from each improvement measure. The relevant contents are provided below as a screen dump for your quick reference.

‘In this study, ShuffleNetV2 is introduced to replace the original backbone network CSP-Darknet53, and deep separable convolution is used instead of traditional convolution to efficiently utilize the feature channels and network capacity while reducing the number of parameters. In the feature fusion network, GhostNet's Ghost module is introduced to reduce the number of parameters and operations of the model by converting the normal convolution operation to generate only some highly differentiated feature maps, and then transforming these feature maps based on cheap linear operations to obtain other similar feature maps. Although the above methods will significantly reduce the number of parameters of the model and speed up the operation speed of the detection network, they will also make the accuracy of detection decrease, so in order to compensate for the loss of accuracy due to the reduction of model parameters, this study introduces a CBAM attention module in front of the detection head. CBAM is an attention mechanism module that combines space and channel to extract positive and effective features in the image, thus improving the accuracy of the network.’

Q2: Please note! You have multiple reference format errors.

Response: We apologize for the reference format errors in the original manuscript. These format errors have been improved.

Sincerely,

The Authors

Reviewer 3 Report

Dear Author,
Thank you for your application with Yolo and speed-up technics.

The presented paper has good potential, but you should distinguish the theoretical idea from the mere Yolo application for deck crew and the use of fishing nets.

You should put in evidence the pros and cons of your speedup method in general and show the application in a second moment. The paper confuses the standard Yolo use with the innovative speed-up technique you introduce. It would be best if you introduced the theoretical aspect and, after some evidence from the experimentation. The application you present is a mere exercise too simple to be published. I suggest you provide more experiments to show the feature of the speed-uo method you introduce.

 Regards

Author Response

Dear Reviewer,

Thank you very much for your time involved in reviewing the manuscript and your very encouraging comments on the merits.

Q1：The presented paper has good potential, but you should distinguish the theoretical idea from the mere Yolo application for deck crew and the use of fishing nets.

Response：Thanks for your great suggestion on improving the accessibility of our manuscript. We have added a brief overview of the overall improvement program of YOLOv5s and the results that can be obtained from each improvement measure. The relevant contents are provided below as a screen dump for your quick reference.

‘In this study, ShuffleNetV2 is introduced to replace the original backbone network CSP-Darknet53, and deep separable convolution is used instead of traditional convolution to efficiently utilize the feature channels and network capacity while reducing the number of parameters. In the feature fusion network, GhostNet's Ghost module is introduced to reduce the number of parameters and operations of the model by converting the normal convolution operation to generate only some highly differentiated feature maps, and then transforming these feature maps based on cheap linear operations to obtain other similar feature maps. Although the above methods will significantly reduce the number of parameters of the model and speed up the operation speed of the detection network, they will also make the accuracy of detection decrease, so in order to compensate for the loss of accuracy due to the reduction of model parameters, this study introduces a CBAM attention module in front of the detection head. CBAM is an attention mechanism module that combines space and channel to extract positive and effective features in the image, thus improving the accuracy of the network.’

Q2: You should put in evidence the pros and cons of your speedup method in general and show the application in a second moment.

Response: Thank you for your suggestion. As suggested by reviewer, we have added the suggested content to the manuscript on page to show the pros and cons of your speedup method in general.

Q3: The paper confuses the standard Yolo use with the innovative speed-up technique you introduce. It would be best if you introduced the theoretical aspect and, after some evidence from the experimentation.

Response: Thank you for the suggestion. We have added the theoretical aspect after the experiment of performance comparison of different models and the ablation experiments.

Q4: The application you present is a mere exercise too simple to be published. I suggest you provide more experiments to show the feature of the speed-uo method you introduce.

Response: Thank you for your precious comments and advice. We have added the feature of the speed-up method after the ablation experiments to show how these methods benefit YOLOv5s.

Sincerely,

The Authors

Round 2

Reviewer 3 Report

I read the revised version of the paper. I asked the authors to put in evidence the improvement concerning the original version of the method. They provide a straightforward answer (lines 148-161) that do not answer the question. I think this paper is a simple exercise of Yolo use and, from my humble point of view, does not match the minimum standard to be published.

The authors do not provide an evaluation of the improvement or do not give a detailed parameter reduction description.

I am sorry that my decision regarding the paper's publication is rejected.

Author Response

Dear Reviewer,

Thank you very much for your time involved in reviewing the manuscript and your very encouraging comments on the merits.

Q1：I asked the authors to put in evidence the improvement concerning the original version of the method. They provide a straightforward answer (lines 148-161) that do not answer the question. 

Response：The content of lines 128-129 is a reply to another reviewer's comment. It is a brief introduction of YOLOv5-SGC.

1. We have added A comparison of ShuffleNetV2 with the original YOLOv5s backbone network, CSP-DarkNet53, illustrates its advantages and disadvantages. (lines 211-216)
2. A description of the ShuffleNetV2 network structure has been added to explain why the modification reduces the network accuracy but increases the running speed and reduces the model size and computation.（lines 248-257）
3. We have added the introduction of GhostModule calculation process and its role.(lines 268-270) and added an explanation and rationale for replacing the C3 module with the CSP_GSC module, describing the rationale that GSConv is more suitable for edge computing devices. (lines 292-306)
4. We have added an introduction to how the CBAM module can be added to the model in this paper, and explained the advantages and disadvantages to the model after the module is added. (lines 343-345)

Q2: The authors do not provide an evaluation of the improvement or do not give a detailed parameter reduction description.

Response:

1. We have added a description of the evaluation method, explaining the role of FPS, parametric quantities, and FLOPs. (lines 368-373)
2. A comparison of confusion matrices for YOLOv5s and YOLOv5-SGC is added to illustrate the comparison of the improved models in terms of number of parameters, computational effort, and accuracy. (lines 406-416)
3. The analysis of ablation experiments was modified, and the effects of adding each module on model speed, number of parameters, calculation volume, and accuracy were elaborated. (lines 442-487)

Sincerely,

The Authors

Round 3

Reviewer 3 Report

Paper already rejected

Author Response

Thank you very much for your time involved in reviewing the manuscript.