ConvDeiT-Tiny: Adding Local Inductive Bias to DeiT-Ti for Enhanced Maize Leaf Disease Classification

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This study proposes a a lightweight hybrid ViT, ConvDeiT-Tiny, for maize leaf disease classification. The model demonstrates excellent classification performance across three datasets, exhibiting both innovation and practical value. Overall, the research objectives are well-defined, the methodology is reasonably designed, and the experimental results are detailed. However, the following issues and shortcomings remain, and the authors are advised to further revise and refine the work:

1. Formula (11) duplicates (10), and the expression for 𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 is incorrect. Additionally, please verify the formula for calculating the F-measure. As described by the authors, using the F1-score would be more accurate, since the F1-score is a special case of the F-measure when β=1, 𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 and 𝑅𝑒𝑐𝑎𝑙 have equal weights.
2. Data augmentation (Gaussian noise, blurring, color dithering) lacks specific parameter details, affecting reproducibility; Although stated to be applied post-splitting, it remains unclear whether this applies “only to the training set.” Additionally, it is recommended to supplement with tables showing the number of instances before and after augmentation across the three datasets. Were all datasets augmented? Furthermore, performance comparisons before and after augmentation should be provided to quantify the contribution of data augmentation.
3. The explanation for N=3 being optimal in the model design is overly simplistic. It is recommended to incorporate a literature review in the discussion section to analyze the reasons for performance degradation when N>3.
4. The dataset name “CD&S” is incorrectly written as “CD%S” in the text; consistency must be maintained.
5. Add cross-dataset performance analysis to the model comparison section. Discuss variations in model performance across different datasets and their causes, while supplementing with additional parameter comparisons to provide interpretable evidence.
6. Subheadings are recommended for the discussion section.

Author Response

Dear Reviewer,

We truly appreciate that you prioritized reviewing our manuscript and, as a result, you were able to provide insightful comments on how to improve our work. After addressing these concerns, we strongly believe that the paper is much better now than it was when we first submitted it. For a point-by-point explanation on how your comments were addressed, please see the corrections matrix attached.

Once again, thank you.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Dear authors,

I am sending my revisions in the attached file.

Comments for author File: Comments.pdf

Author Response

Dear Reviewer,

We truly appreciate that you prioritized reviewing our manuscript and, as a result, you were able to provide insightful comments on how to improve our work. After addressing these concerns, we strongly believe that the paper is much better now than it was when we first submitted it. For a point-by-point explanation on how your comments were addressed, please see the corrections matrix attached.

Once again, thank you.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

Dear authors,

Many thanks to the authors for taking into consideration all the suggestions provided, but in the new R2 version, references appear in the results section. In this section, the authors should only describe the tables or figures of their results, while in the Discussion section they should compare their results with those of other authors and add those references here, thus creating a robust discussion.

Author Response

Dear Reviewer 2,

Thank you for prioritizing reviewing the revised version of our manuscript. We appreciate your great attention to detail and your desire for us to improve the paper to the highest quality possible before it is published. We have addressed the two concerns you raised in the second round of review, and we believe that this has strengthened the paper even further.

For a point-by-point explanation on how your comments were addressed, please see the corrections matrix in the attached document.

Author Response File: Author Response.pdf