Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear authors:

I appreciate the opportunity to review your work, which focuses on an endangered species (Xishuangbanna cucumber) with real difficulties in seed viability assessment, which represents a significant contribution to genetic conservation. This non-destructive technique in this context is appropriate and relevant, offering a potentially applicable alternative in germplasm banks. Evaluating multiple algorithms (KNN, LogitBoost) and preprocessing techniques (MSC, SNV, FD, SD, L2NN) demonstrates a rigorous effort in computational analysis. However, we also identified significant weaknesses that need to be addressed. Below are several observations:

1. The study focuses exclusively on a single species and uses widely known models (KNN, LogitBoost) without introducing technical innovations or a novel conceptual framework. We recommend discussing the implications of applying this approach to other species and justifying the methodological choice over more recent models. 

2. The contrast with state-of-the-art deep learning algorithms such as attention networks, transformers, or deep ensembles, which are now standard in spectral data analysis, is omitted. Including these comparisons or mentioning them would significantly strengthen the discussion.

3. Relevant and recent references that demonstrate more robust approaches to hyperspectral image classification using deep learning are essential to add.

4. Section 2.5 does not argue whether the selected lineages adequately represent the species' genetic diversity. It is recommended that this selection be justified or, failing that, that the set analyzed be expanded.

5. The abstract presents redundancies and does not communicate the problem, methods, key results, or contribution. We suggest reformulating it with a more direct structure: problem – method – results – impact.

6. The discussion lacks concrete scenarios on how the results could be implemented in genebanks or other species. Expanding this section to project its usefulness beyond the case study would be valuable.

7. Some graphs, such as Figures 2 and 6, require more descriptive legends to facilitate their interpretation without the need for extensive revision of the text.

8. The current ones are too generic. We recommend including more specific terms such as machine learning, seed vigor, germplasm conservation, and hyperspectral imaging.

9. Repetitions and a sequence that mixes results with discussion are identified, diluting the critical analysis. Reorganizing this section to highlight implications, limitations, and relationships with recent studies will improve its impact.

10. The results are excessively long and repetitive. The key findings should be summarized by section, and redundancies should be avoided.

11. Although the conclusions are correct, no specific lines of future work are proposed, nor is the limitation of not using Deep Learning methods explicitly acknowledged.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

In this work the authors explore the implementation of machine learning (ML) algorithms to predict variability of Xishuangbanna cucumber seeds. The authors effectively used hyperspectral imaging technology along with ML and they found that L2NN-KNN model had the highest model accuracy. Experimental results are then compared, and the paper provides good details about the application of different ML models. However, some aspects require clarification, and the figures need further polishing to improve presentation quality. Here some comments for the authors to improve the manuscript:

- In the abstract there is a sentence that could be rephrased: "Therefore, in order to realize the rapid and nondestructive detection of seed viability of Xishuangbanna cucumber, Xishuangbanna cucumber seeds with different viability levels were specially selected as the research object, and their hyperspectral images were collected in the wavelength range of 400-1000nm"

- I suggest adding the following relevant work about the use of hyperspectral imaging and ML to the state-of-the-art review: 

https://doi.org/10.3390/s24020344

- Including an image of the seeds or the experimental setup would significantly enhance the reader's understanding of the study.

- In most of the Figures the font size is too small and thus hindering the readability, especially for labels in fig. 2A-B. Additionally, the inconsistent panel sizes, as in Figure 5, should be standardized along with uniform font sizes throughout the figures to improve the presentation's clarity and visual consistency.

- Figure 5e represents the diagram of the CARS screening process, but it is not clearly readable or appropriately labeled, therefore it is rather confusiong when the autors explained or tried to comment the figure in the text. 

- In section 3.5 feature band extraction with SPA algorithm is discussed, the authors state that when the number of selected bands is 12, RMSE reaches a minimum. Could you please provide details regarding the precision of the wavelenght values that you have reported for these selected bands?

- Section 4 reads more like a conclusion rather than a discussion of the results obtained by the experimental study. To improve clarity and flow, I suggest adding a distinct concluding section and restructuring the discusion section. This could involve first discussing the obtained results in comparison with existing state-of-the-art findings (Discussion), followed by a separate concluding section to summarize the key takeaways (Conclusion).

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The inclusion of Figure 1 has improved the paper; however, further refinement is needed. The lack of specific labels on the system schematic in Figure 1 makes it difficult to understand the components beyond the computer and monitor.

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx