An Improved YOLOv5 Algorithm for Tyre Defect Detection
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsWe appreciate the authors for contributing to the research on the detection of tyre defects.
The contributions of this study are as follows.
1. Level 1: Contributed DySneakConv and AIFI to the backbone of the YOLOv5 framework, as mentioned in Section 2.2.1.
2. Level2: Authors developed the CARAFE and replaced the upsample at the Neck of the YOLOv5 framework in Section 2.2.
3. Equations 1–4 were used to estimate the positions of the fractional coordinates using bilinear interpolation on a discrete grid of integer coordinates. The same is shown in Figure 3.
4. Complete Framework presented in section 3 with lightweight detection algorithm. The relevant results are presented in Section 4.
5. Among the results, figure 11 presents the novel contribution of the works with tyre defect detection applications. Tables 1 to 3 support the varied permutations of DySneakConv+ AIFI+ CARAFE.
Queries:
1. Why authors do not replace all parts of the Bottleneck CSP with DySneakConv, it limited up to level-1 fold only, as shown in Figure 7.
2. In Figure 4, the general structure of AIFI, inputs applied as S5, does not provide clear images.
3. 2.2.1 are repeated, such formatting errors were repeated.
4. If feasible, the authors provide a comparison of the efficacy of the proposed approach with other methods.
5. The introduction is well presented, but a conclusion reframe with competitive results and outcomes is required.
Author Response
Thank you very much for taking the time to review this manuscript. Detailed responses are in the PDF.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsDear Authors
Thank you very much for your work. This is a well-written article with an interesting approach about advanced computer vision models applied to tyre defects. I’m willing to reconsider after major revision. Please find detailed comments below.
Major concerns:
(1) The novelty of the proposed framework is limited, detection by deep neural networks and transfer learning approaches based on pre-trained models are not new. The novelty in this study is using such approaches for tyre defects use case with all its challenges.
(2) The proposed enhanced YOLOv5 model’s files should be shared using e.g. zenodo, a DOI and a license. This is the main point of open science.
(3) In 2020, Ultralytics unveiled YOLOv5, in 2023 unveiled YOLOv8, and in 2024 YOLOv9. The authors proposal is a YOLOv5 enhacement and not a YOLOv9 enhacement, not even a YOLOv8 enhacement.
(4) Although the authors meontion that the dataset was randomly divided in a training set and a test set, the results reported in the abstract and also in table 1, table 2, table 3, table 4 are only the training results. Inference results using the test set should also be provided, these are the ones that should be reported in the abstract.
(5) There is not true discussion section in the paper, the so-called "discussion" section should be renamed: 1) 4.1 and 4.2 should go to a new section called methodology; 2) 4.3, 4.4 and 4.5 should go to a new section called results (4.5 and 4.6 should be only one section). A the true discussion section must be included, where: 1) the authors compare results with other authors that addressed tyres's defect detection; 2) explain why YOLOv8 or YOLOv9 network's improvement were not also taken into consideration; 3) include some limitations in the discussion, for instances the proposed network represent an improvement over YOLO's only for this subset of images related to tyres' x-rays and not in general.
Minor concerns:
(1) Throughout the paper different mAP figures are compared. As mAP is stated in percentage, differences between percentages must be stated in percentage points (pp) as the unit for the arithmetic difference between two percentages and not percentage increase or decrease in the actual quantity. For example, if the percentage of correct detections increases from 80% to 85%, the change is 5 pp, not 5%.
(2) include a link to the open-source dataset used.
(3) Figures 11, 12 and 13 should be provided as suplementary material.
Author Response
Thank you very much for taking the time to review this manuscript. Detailed responses are in the PDF.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe paper presents an improved YOLOv5m algorithm for tyre defect detection. However, some problems should be addressed as follows:
1. Why did the author improve YOLOV5 instead of YOLOV9, because YOLOV9 now has good performance.
2. The experimental comparison in the experimental section is not sufficient and lacks comparative analysis with the latest framework.
Comments on the Quality of English Language
normal
Author Response
Thank you very much for taking the time to review this manuscript. Detailed responses are in the PDF.
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsKindly refer to attached file for my comments.
Comments for author File: Comments.pdf
Ok
Author Response
Thank you very much for taking the time to review this manuscript. Detailed responses are in the PDF.
Author Response File: Author Response.pdf
Reviewer 5 Report
Comments and Suggestions for AuthorsThis study on the Improved YOLOv5 Algorithm has achieved better Tyre Defect Detection. I have some suggestions for the authors to consider:
1. It is recommended that the differences between this study and the referenced literature be explained in the literature review section, possibly by creating a table for comparison.
2. In Chapter 3, numerous references are still found in the results section. I believe these should be explained in Chapter 2. Chapter 3 should focus on presenting the research results.
3. For Figure 10, each figure should be individually explained. It is suggested that different colored boxes be used for explanation to enhance readability.
4. The categories in Figures 12 and 13 should be explained individually with formatting adjustments to improve readability.
5. Including additional relevant literature and explaining the differences is suggested. Electronics would be a good category choice for this purpose. In light of this, please provide 2-4 references for the authors to consider.
(1) Using image analysis techniques to enhance TFT-LCD manufacturing yield through timely detection of PI alignment layer defects
(2) Detection and Analysis of Aircraft Composite Material Structures Using UAV
(3) Monitoring of defects of a photovoltaic power plant using a drone.
(4) A study on nitrogen oxide emission prediction in Taichung thermal power plant using artificial intelligence (AI) model
Author Response
Thank you very much for taking the time to review this manuscript. Detailed responses are in the PDF.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsDear Authors,
I'm happy with the revised version. Details regarding data are still missing and these are very important for readers and further use of your research: 1) Data descriptor information; 2) Data availability statement. Notice that a DOI should be used for data and not a link to a google drive. Check zenodo or similar free platforms. Guidelines are available here: https://0-www-mdpi-com.brum.beds.ac.uk/journal/data/instructions
Author Response
Thank you very much for taking the time to review this manuscript. Detailed responses are in the PDF.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe paper has addressed all my questions.
Comments on the Quality of English Languageminor
Author Response
Thank you very much for taking the time to review this manuscript. Detailed responses are in the PDF.
Author Response File: Author Response.pdf
Reviewer 5 Report
Comments and Suggestions for AuthorsMost articles have been revised, but a good article should have 25-35 references and be acceptable after minor repairs.
Author Response
Thank you very much for taking the time to review this manuscript. Detailed responses are in the PDF.
Author Response File: Author Response.pdf
