An Appearance Defect Detection Method for Cigarettes Based on C-CenterNet

Round 1

Reviewer 1 Report

The paper presents a Cigarette defect detection method based on CNN and Classification. The used techniques and technologies are well known, the problem statement is well presented and the experimental results demonstrate the benefits of the suggested method. Therefore the manuscript can be accepted for publication. 

The paper presents a convolutional neural network based defect detection method focusing on the tabaco industry. The problem is well presented and the existing solution are reviewed. 

The authors use the Restnet50 and tailored the network for the application.  Some layers of the original network was removed or modified according to the application are. The used techniques are already known and they are applied properly. 

The evalution of the defect detection algorithm was based on the precision and the recall. The training was performed on a 2000 images. The data set was split into 6:2:2 ratio. The description of the data set could be clarified. Figure 9 is hard to interpret. It is not clear why there were exactly 10.000 images and why the distribution of the different defects are the same. Is it possible to have more than one deffect on an image?

Experimental results clearly shows the suggested CCenterNet method superior to the existing solutions. Only the YOLO5 was faster than the CCenterNet. Considering the real life application of this algorithm, what does this speed difference mean? Does it limits the applicability of the CCenterNet? 

To sum up, the paper presents a novel application of CNN and classification focused on cigarette defect detection. The presented method was tested and it performed better than the existing solutions. Therefore, the acceptance of the paper is suggested.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The article "An Appearance Defect Detection Method for Cigarettes Based on C-CenterNet" deserves the reviewer's best appreciation and attention.

This study provides an interesting insight into a current and pertinent subject.

It would be interesting to have a more recent bibliography

There must be reinforcement:

What is the purpose of the study?

What is the research of the study?

What brought us back to the study?

Who or what are the contributors?

There should be a strengthening of the research question

I miss the theories and practices and the study

Hope this helps,

No other subject,

The hug

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors have presented a defect detection method for cigarettes based on the improved CenterNet. The manuscript is well written and the improved network is tested and validated using experimental results. The manuscript can be accepted after correcting following modifications.

-        Kindly add more related works about cigarette defect detection in the introduction section.

-        In line 58 in the introduction section: “In terms of cigarette appearance defect detection, the current research almost entirely uses traditional digital image processing methods to solve the problem”, what is the meaning of this sentence by the authors. It is claimed about the presented paper or other cited researches; it should be clarified.

-        The texts in Figures 1 and 2 cannot be read. The quality of this figure should be increased.

-        Some parameters in text should be aligned, such as the parameters in line 112, 113, 116, and etc.  

-        The equations, which are not extracted by authors should be cited, for example the loss function of CenterNet and etc.

-        Ref [18] is not cited in the text.

-        showing the detection results of cigarettes in different background environments may be helpful.

-        Kindly indicate the parameters of presented CenterNet model.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

.

Author Response

Dear Reviewer:

Thank you for your arduous work.

Sincerely yours,

Guowu Yuan

Author Response File: Author Response.pdf

Reviewer 3 Report

The revised version of the manuscript is improved and the manuscript can now be accepted after minor modification.

-        Correct typo in lines 133.

Author Response

Dear Reviewer:

We would like to thank you for your careful reading, helpful comments, and constructive suggestions, which have significantly improved the presentation of our manuscript.

Point 1：

Correct typo in lines 133.

Response 1：

We have checked for errors on line 133 and made corrections, changing "25]" to "[25]".

At last, thank you for your arduous work and instructive advice.

Sincerely yours,

Guowu Yuan

School of Information Science and Engineering, Yunnan University

Author Response File: Author Response.pdf