Improved Apple Fruit Target Recognition Method Based on YOLOv7 Model

Round 1

Reviewer 1 Report

This academic study proposes an algorithm based on YOLOv7 for improved apple fruit detection, aiming to enhance accuracy in scenarios with high fruit density and overlapping instances. The research suggests a preprocessing algorithm for handling overlapping regions, providing a parametric fusion with the base network of YOLOv7 using the MobileOne module and aiming to reduce network computations. Subsequently, the SPPCSPS module is enhanced and modified to transform the image feature fusion from a serial channel to a parallel channel, increasing the speed. The results are compared with other YOLO algorithms. The authors state that the accuracy of the improved YOLOv7 algorithms is higher compared to other control YOLO algorithms. This study serves as a reference for other academic works in terms of content and subject matter. Additionally, the study appears to be written in an academic style and language. However, for the publication of this study in a journal, the authors need to consider some necessary revisions. These revisions can be listed as follows:

• The Introduction section lacks sufficient information about previous academic studies that employed the YOLO algorithm for apple fruit recognition. An example study can be reviewed below: https://www.mdpi.com/2072-4292/15/6/1516
• In the Introduction section, it is advisable to provide a preliminary overview for readers by presenting the previous research on apple fruit recognition and the results of those studies, along with an introduction to the current study.
• The abbreviation CBAM, mentioned for the first time in line 62, needs to be fully explained. Therefore, it is necessary to review other instances in the remaining part of the study where such abbreviations are not followed by a comprehensive explanation. For instance, in line 227, the term CIoU is used.
• The dataset lacks information about the raw apple images and whether any preprocessing was applied. Including an example image illustrating the preprocessing steps would be informative for readers.
• Section 3 should include a block diagram of the study and a brief paragraph explaining the stages of the study.
• Table 2 should be presented on a single page. Currently, it spans across two separate pages.
• The Results section should be expanded, including suggestions for future research.

Author Response

Thank you for your suggestions. Please see the attachment for our reply.

Author Response File: Author Response.docx

Reviewer 2 Report

This paper discusses the application of an improved YOLOv7 network to detect apple fruit. Here are some points I’d like to highlight.

1.       The Introduction proposes a brief state of the art, which does not account for several novel applications of YOLO to the field of crop detection. Please find enclosed the following relevant citations to be included to improve this section

a.       https://doi.org/10.1016/j.biosystemseng.2021.08.015

b.       https://doi.org/10.3390/agronomy12020319

c.       https://doi.org/10.1016/j.compag.2023.107757

2.       The first paragraph of section 2.2 lacks references and is hard to read. Please improve it.

3.       Why do the authors compare YOLOv7 with YOLO (I assume the first version) in section 3.1? A proper comparison should be with its nearest predecessors, such as YOLOv6 and YOLOv5.

4.       Figures 1 and 2 are hard to read. Furthermore, the authors should improve the description of the difference between the two structures to improve the readability of the paper.

5.       Overall, the authors propose a specific configuration for the YOLOv7 base network using existing blocks and functions. Does the work in this configuration introduce the novelties? If so, the authors should stress why each of these blocks was introduced in the context of the application scenario.

6.       Why did the values of the provided metrics fluctuate within the first 30 iterations?

7.       Table 2 appears to describe an ablation test. Please highlight directly within the table which configuration achieved better results for each metric to improve readability. Also, why the latest row does not have any mark for SPPFCSPC?

8.       Figures 9 and 10 spans two pages. Fit it to a single page to improve readability.

9.       It appears that the original YOLOv7 was used after training on the proposed dataset in the evaluation in Table 3. Please specify if the training was performed from scratch or using transfer learning.

10.   From Table 4, YOLOv3 outperforms YOLOv5. Did the authors provide a plausible explanation for that? Which version of YOLOv5 did the authors use (nano, small, medium, large, extra large)?

Overall, the paper can be considered for publication after these issues are fixed, hence I suggest a major revision.

 

 

1.       The narrative uses forms that are not suited for a scientific paper.

a.       Please use the third plural person instead of the first plural person.

b.       Please do not use possessives (e.g., world’s).

 

c.       Please check tenses and typos.

Author Response

Thank you for your suggestions. Please see the attachment for our reply.

Author Response File: Author Response.docx

Reviewer 3 Report

please see file.

Comments for author File: Comments.pdf

can be improved by editing

Author Response

Thank you for your suggestions. Please see the attachment for our reply.

Author Response File: Author Response.docx

Reviewer 4 Report

Dear Authors,

      His article "Improved Apple Fruit Target Recognition Method Based on YOLOv7 Model" addresses a relevant topic that still needs proper resolution. In order to contribute to the improvement of the writing of your manuscript, I present observations in the comments of the digital file and request your attention to the following comments:

1) Line 58: Recently, methods that employ neural networks with deep learning have gained relevance. In this context, Altaheri et al.... Thus, you highlight the evolution of methods with innovative techniques.

2) The application of the fruit detection results in Chapter 1 was lacking.

3) Line 80: Highlight the study's objectives at this point.

4) Lines 96 to 99: detail the image acquisition method. It would be interesting to illustrate this step by presenting a figure.

5) Present the characteristics adopted for image acquisition: focal length, f stop, ISO, and other configurable camera parameters.

6) Line 101: “As the camera's resolution increases, the Apple image resolution acquired increases”. That much is evident. In this case, did you use the camera's native resolution and resampled the images later?

7) Line 124: inform the characteristics of the hardware resources.

8) Line 199: “This replacement improves the speed and enhances richer feature information”. How does it enhance richer features?

9) References for formulas 1 to 4?

10) Table 1: highlight the highest values using bold.

11) Line 366: “This study selects five conditions: smooth close-range lighting, close-range reverse lighting, smooth close-range lighting, close-range reverse lighting, and large-field scene.” Are the first, and second similar to the third and fourth study conditions? Please detail the criteria and requirements.

12) In the results, it would be interesting, in addition to comparing the results obtained from the proposed algorithm with traditional algorithms, to bring some comparison with the results obtained by the authors cited in the bibliographical references.

13) Lines 434 to 447: the conclusions end up repeating information already presented in the text. It would be interesting to point out that the proposed objectives were achieved. What would be the recommendations for future studies?

      I conclude by congratulating them for the study carried out and for the presented version of the article.

 

Respectfully,

 

Comments for author File: Comments.pdf

    Attention to long paragraphs as they make reading confusing. Check the use of punctuation, especially the comma. Pay attention to verb agreement.

 

Author Response

Thank you for your suggestions. Please see the attachment for our reply.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The authors successfully fixed the highlighted issues. The paper can be considered for publication.

Reviewer 3 Report

The authors have solved all my concerns and issues, thanks！

English is fine，can be improved by further carefull revise.

Reviewer 4 Report

Dear Authors,

    The second version of your article: "Improved Apple Fruit Target Recognition Method Based on the YOLOv7 Model," presents several changes in the writing and structure of the manuscript. Comparing the second with the first version and aided by its cover letter, I could see that all my suggestions were implemented or duly justified.
From that analysis, my recommendation is to accept your article for publication.

      I conclude my comments by congratulating them for the research and article presented.

 

 

Respectfully,

 

There are a few minor punctuation and verbal agreement issues with the text.