Utilization of Machine Learning and Hyperspectral Imaging Technologies for Classifying Coated Maize Seed Vigor: A Case Study on the Assessment of Seed DNA Repair Capability

Reviewer 1

The manuscript “Utilization of machine learning and hyperspectral imaging technologies for classifying coated maize seed vigor: A case study on the assessment of seed DNA repair capability” proposes a non-destructive technique for the assessment of coated maize seed vigor.

The study is interesting, and the experimental design is well planned.

Minor changes are suggested:

Comments 1: Lines 125-125: Enter which harvest times, the number of seed producers, and the number and location of production areas.

Response 1: Thank you for your valuable comments. We agree with this comment. Therefore, the research methodology has been updated.

Comments 2: Line 126: Why was a film-coating technique used?

Response 2: The utilization of film coating techniques is on rise due to their cost-effectiveness and efficiency in the production of commercial maize seeds.

Comments 3: Line 138: Figure 4 should be inserted here as figure 2.

Response 3: We sincerely thank the reviewer for constructive criticisms and good suggestions. The citation for this figure has been removed in order to maintain the order of the images without disrupting the primary content of the article.

Comments 4: Line 202: It would be appropriate to provide more information about the CARS technique.

Response 4: We have, accordingly, added more explanation of the CARS technique to emphasize this point. CARS leverages partial least squares regression and discriminant analysis to effectively pinpoint key wavelengths within data. Please refer to line 214 on page 5.

Comments 5: Lines 212-214: With which software were oversampling techniques used?

Response 5: We employed the imbalanced-learn python package for both oversampling techniques used in our study. We have also added this in the revised manuscript (line 255 on page 6).

Comments 6: Figure 3: Increase the size of the x-axis. The current version is not readable.

Response 6: We sincerely thank the reviewer for valuable comments. The figure has been revised in accordance with the recommendations.

Comments 7: Line 312: Indicate the value or range of values of the short time frame.

Response 7: Thank reviewer for sharp suggestions. The value or range of the time frame has been further elucidated (line 337 on page 12).

Comments 8: Lines 312-313: This sentence needs further clarification in both the introduction and the discussion.

Response 8: Thank reviewer for the useful suggestions. The time-based comparison between the original method and our novel approach has been further explored in the introduction and discussion.

Comments 9: Fix references with correct MDPI Journals formatting and eliminate double numbering.

Response 9: Please accept our appreciation for your exceptional recommendations. Double-numbering has been eliminated, and the references have been rectified.

Reviewer 2

The paper presents an interesting study on the use of machine learning and hyperspectral imaging technologies for classifying the vigor of coated maize seeds. The topic is highly relevant, given the increasing demand for efficient and non-destructive methods in agricultural practices. The authors have made a good contribution by focusing on the practical application of these technologies, which aligns well with current needs in the field.

However, while the study's premise is promising, the results provided are limited and do not fully explore the potential of the proposed methods. The analysis would benefit from additional experiments to enhance the robustness of the findings. For example, expanding the dataset or incorporating more diverse seed conditions could provide a deeper understanding and validation of the proposed framework. This would not only strengthen the conclusions but also offer more comprehensive insights for real-world applications.

Additionally, the paper could improve in several areas:

Comments 1: Methodological Depth: The current study provides an overview of the methods used but lacks a detailed exploration of the rationale behind the choice of specific techniques. A deeper discussion on why certain preprocessing and machine learning models were selected, compared to others, would add valuable context. The paper could include more detailed statistical analysis and validation metrics to support the results. For example, the use of cross-validation, confusion matrices, or additional performance metrics (e.g., precision, recall, F1 score) would provide a more robust evaluation of the machine learning models.

Response 1: Thank you for pointing this out. We have expanded our discussion on the selection of pretreatment analysis and machine learning models. For the pretreatment analysis, we chose SNV and MSC, two widely used techniques in spectroscopy to enhance data quality and improve model performance. Please see line 205 on page 5 for additional details. Regarding ML models, we opted for popular algorithms known for their ability to handle high-dimensional data like ours. We have included further discussion at line 237 on page 6 in the revised manuscript. Additionally, we included extra validation metrics, including accuracy, precision, and F1-score in Table 2.

Comments 2:  Incorporating an analysis of uncertainty or sensitivity could add depth to the findings. Discussing the reliability of the results, considering the potential variations in seed conditions or imaging parameters, would demonstrate a comprehensive understanding of the experimental limitations.

Response 2: We sincerely thank the reviewer for valuable comments. A discussion of the reliability of results that could be affected by the sample seeds used in model training has been provided between lines 397 – 402, while the limitations and prospects for further research on these algorithms are explored in lines 414 – 419. In addition, hyperspectral image acquisition at wavelengths outside the study is also discussed with references to appropriate publications.

Comments 3: Practical Implications: The paper briefly touches on the potential practical applications of the study but could do more to explore the implications for industry and future research directions. Expanding this discussion would provide a clearer understanding of the study’s impact and relevance.

Response 3: We sincerely thank the reviewer for constructive criticisms and insightful comments. The study's potential practical uses for industry and future research directions have been described on in lines 407 – 414.

Comments 4: In conclusion, while the paper is well-conceived and tackles an important topic, it would significantly benefit from a more comprehensive analysis and discussion. Expanding the experiments and providing more detailed comparisons with existing literature would enhance the paper's contribution to the field.

Response 4: We value the reviewer's ideas for a more balanced and nuanced consideration of some of the manuscript's concerns. Several statements that we made were more ambiguous than intended. The manuscript has been revised according to the concerns for resubmission and publication.

Reviewer 3

This is a meaningful study of corn seed activity detection using hyperspectral imaging and machine learning techniques to categorize corn seed activity and ensure consistent corn seed activity within the same seed lot. I have some comments on the textual presentation, image formatting, article structure, citation additions and deletions in this study：

Comments 1: In lines 30 and 43, add relevant references to the description of the article.

Response 1: Thank you for pointing this out. We agree with this comment. The relevant references have been cited.

Comments 2: In line 98, the introduction to experimental design could be placed in chapter II.

Response 2: Thank you for giving us the opportunity to explain. The experimental design was fully described in chapter II (Materials and Methods) to enable reproduction and further development of the published results, including appropriate references. However, it is beneficial to provide a brief overview of the experimental design in the Introduction (chapter I) to guide the reader to the subsequent chapter. This paragraph should, therefore, be maintained.

Comments 3:  In line 197, the diagram is blurred, please redesign the diagram and reflect the main elements of the diagram in the text.

Response 3: We appreciate the reviewer's insightful recommendations. We have, accordingly, revised the diagram to emphasize this point.

Comments 4:  In line 207, please describe the reasonableness of the data distribution ratio or add relevant references.

Response 4: A small proportion of 10% is held-out for testing purposes. It is sufficient for a final evaluation, as it provides a representative sample of unseen data. The remaining data is then divided into training and validation sets. Given the relatively large number of spectral features, the model can potentially be at risk of overfitting. To mitigate this, we intentionally set a slightly large validation set (15%) compared to the testing set. This allows for effective hyperparameter tuning and helps prevent overfitting by providing an unbiased estimate of the model’s generalization performance. Please refer to line 224 of page 6 in the revised manuscript.

Comments 5: In line 121, it is proposed to add a description of the principles and operational processes of the machine learning techniques used in the research process.

Response 5: We have included deeper explanations regarding SVM and ELDA algorithms, See line 237 on page 6.

Comments 6: The pictures related to the article are blurred, please adjust the format of the pictures according to the requirements of the journal.

Response 6: We sincerely thank the reviewer for valuable comments. The figure has been revised in accordance with the recommendations.

Comments 7: In line 303, in the course of the discussion, data on the results of the study could be added to the description to support the discussion.

Response 7: We gratefully acknowledge the reviewer's excellent recommendations. In responding to the reviewer's comments and concerns, the discussion has been updated in the revised version of the manuscript in response to the reviewer's comments.

Comments 8: In line 393, in the description of the conclusions, relevant data from the analysis and conclusions could be added to make the relevant conclusions more convincing.

Response 8: We sincerely thank the review for valuable comments. The conclusion has been updated in the revised version of the manuscript in response to the reviewer's comments.