Grading Algorithm for Orah Sorting Line Based on Improved ShuffleNet V2
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe paper effectively frames the problem of inefficient and inaccurate manual Orah grading in China, highlighting the need for automated solutions. The proposed solution combines machine vision and deep learning, specifically improving ShuffleNetV2 with the Mish activation function and ECA attention module. This topic is highly relevant and aligns well with the scope of the Applied Sciences Journal. However, there are some concerns regarding your manuscript that need to be addressed as follows:
1) The contribution is significant for agricultural automation, but the paper could better emphasize how these improvements surpass existing solutions (e.g., quantitative comparisons with other attention mechanisms).
2) Some citations lack diversity (e.g., predominance of Chinese theses). Authenticity and accuracy appear sound, but including more international studies could strengthen the background.
3) The research is well-designed, but the provided data is not always sufficient to fully support the conclusions. More details on experiment reproducibility, including datasets and parameter configurations, would improve transparency and reliability.
4) The paper follows a logical structure but is excessively wordy and could be more concise. Minor grammatical and stylistic refinements would improve readability. Additionally, some figures and tables could be enhanced for better clarity.
5) The paper lacks a dedicated discussion on testability and methodological robustness. Addressing potential flaws, missing controls, and alternative explanations would improve the credibility of the findings.
6) The technical foundation of the paper is strong, but the validation process could be more rigorous. There is limited benchmarking against existing techniques, and further comparative analysis would help substantiate the claims. Additionally, a more detailed discussion on reproducibility and potential sources of error would be beneficial.
7) There are no major flaws, but the diameter algorithm’s reliance on fixed thresholds (e.g., Dmin=40, Dmax=200) may not generalize to all fruit sizes.
8) The paper lacks details on hyperparameters (e.g., learning rate schedules) and hardware specifications (e.g., camera models). Reproducibility could be improved by open-sourcing the code and dataset.
Comments on the Quality of English LanguageThe English could be improved to more clearly express the research.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe paper entitled "Sorting algorithm for Orah sorting line based on improved ShuffleNetV2" presents a relevant topic in the field of agricultural automation, and it makes a significant contribution through the development of a lightweight deep learning model, adapted for industrial Orah sorting lines. However, in its current form, the paper requires major revision before it can be considered for publication. Below are my comments and specific suggestions for improvements to be made to the paper:
1 - There is a lack of information on what operational constraints and limitations the proposed system has. The authors should consider addressing sensitivity to lighting variations or fruit soiling; Fragility of Grade B classification (accuracy of only 76.9%); Generalizability to different Orah cultivars or environmental conditions.
3 - A comparison with conventional image processing approaches (thresholds, morphological techniques, etc.) is missing to better position the contribution of the proposed work.
4 - Several sections of the article contain repetitive phrases, long sentences, and strange constructions. A professional English review is recommended.
5 - Authors should include references or parameters from other classification studies of fruits with similar visual characteristics (e.g. oranges, tangerines), in addition to apples.
6 - It remains to be specified whether the augmentation techniques were applied equally in all classes or whether a balance was necessary.
7 - The article lacks information about implementation potential (e.g. embedded devices, mobile integration) as the model is lightweight.
8 - There is a lack of clarification on whether the heat maps were generated using Grad-CAM or another technique.
The English could be improved to more clearly express the research.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsNo additional comments
Author Response
Thank you for your review. Although no specific comments were provided, we have made further improvements to the manuscript. In particular, we refined the research design, clarified the methodology, and improved the overall clarity of the language.
Reviewer 2 Report
Comments and Suggestions for AuthorsThe revised article shows significant improvement, and the authors responded appropriately to the review comments. However, there are still some points that need to be improved, namely:
1 - Although the authors mention that conventional methods such as thresholding and morphological operations were ineffective, the inclusion of a small comparative table (even with poor results) would help demonstrate the superiority of the system created.
2 - The article lacks information about the limitations of transferring the model to different cultivars or environments.
3 - The confusion matrix shows relatively lower performance for Grade B (76.9%) and although some analysis has been provided, suggestions for future improvements in feature representation or data balance should be inserted in the paper.
4 - There are still long sentences and the English needs to be reviewed.
The English could be improved to more clearly express the research.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
