A New Knowledge-Distillation-Based Method for Detecting Conveyor Belt Defects
Round 1
Reviewer 1 Report
In my opinion, the article presents an original approach to the topic of using machine learning and deep learning methods to assess the state of transmission belts. The approach presented by the authors is new and is not based on well-developed algorithms based only on neural networks. I did not find any weaknesses in the text. The manuscript is clearly and fully written. I strongly support its publication.
Author Response
Thank you for your recognition of our research. We have made further additions and revisions to the article based on the comments of other reviewers and invite you to review it.
Reviewer 2 Report
Review of the article Applsci-1919071
A new knowledge distillation-base method for detecting conveyor belt defects.
The authors propose a data augmentation technic based on a Gan and a copy-pasting method for increasing the number of example images for training. After that, a YOLOv5 net is reduced to generate two mini-networks, the YOLOv5n and the YOLOv5n-slim. The first one has better efficiency measurements than the second one, which is smaller than the first one.
In my opinion, the authors propose a novel method for the data augmentation problem and the smaller networks' generation using a distillation approach based on other published articles. The results are very encouraging.
To improve the article, I propose to say something about the degree of improvement by using the data augmentation technique. To this end, I suggest comparing the networks trained with the original images and with the produced by the GAN and copy-pasting method, so the readers can better understand the advantages of the proposed method, a kind of ablation study concerning the data.
Also, it seems that the number of 7276 generated images is incorrect because you have included the 1533 images that were taken directly from the conveyor belt, and generated means only the generated by your augmentation strategy, is this asseveration correct?
Several sentences need to be improved. Here are some of them; please correct them: its basis. the network; as shown in Figure 10 shown.; this paper we compared with other YOLO….
Improve Fig. 12; the colors of the points are hard to recognize. You can use other characters, not only points, to construct the graph.
Author Response
Thank you very much for your recognition of our research. Your suggestions have been very helpful in improving our paper, and we have addressed each of the issues you raised. Please see the attachment for more details.
Author Response File: 
Author Response.docx
Reviewer 3 Report
The authors have used an innovative method for object detection and localization to solve the defect detection problem in conveyor belts. The technique has some real-life applicability, but the authors haven’t discussed how they propose to deploy their models in real life.
1. Section 2.1 is almost redundant and already present in the literature.
2. Literature review should be presented in a tabular fashion to complement the easier understanding of different works.
3. A section about the dataset description should be presented, and the frequency distribution of all the classes of data should be given.
4. The bibliography established the credibility of the work, authors are advised to consider the latest papers on the subject.
5. A comparative analysis of the proposed algorithms to other works on similar datasets should be presented.
6. Authors are advised to compare their work with other research works published in the area.
7. Authors are advised to comment on the performance of their proposed algorithm.
8. Authors are advised to discuss the future research opportunities of the work.
Author Response
Thank you very much for your recognition of our research. Your suggestions have been very helpful in improving our paper, and we have addressed each of the issues you raised. Please see the attachment for more details.
Author Response File: Author Response.docx
Round 2
Reviewer 3 Report
Accept
