Deep Learning-Based Model for Effective Classification of Ziziphus jujuba Using RGB Images
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsDear Editor,
I am writing to you as a reviewer of the manuscript titled "Deep Learning-Based Model for Effective Classification of Ziziphus jujuba Mill. Using RGB Images" which was submitted to AgriEngineering with the manuscript ID agriengineering-3310215.
This article presents a significant contribution to the field of medicinal plant classification by leveraging advanced deep learning methodologies. The authors successfully develop and evaluate three convolutional neural network (CNN) architectures—ResNet-50, Inception-v3, and DenseNet-121—demonstrating their applicability in distinguishing between species of the Ziziphus genus based on RGB images of seeds. The study addresses a pertinent issue in the herbal market, namely the challenges posed by the genetic and physical similarities among medicinal materials, which can lead to counterfeiting and quality assurance difficulties. Some comments for each section are as follows:
- Abstract:
The abstract conveys a clear and compelling message about the potential impact of the research on quality assurance in the herbal market, while setting the stage for the detailed study that follows.
- Introduction:
The incorporation of a review of existing literature highlights advancements in both traditional and modern classification methods, supporting the rationale for using deep learning techniques. Moreover, the focus on Ziziphus jujuba Mill. and its potential classification challenges is relevant, as it addresses a gap in current methods. The stated objective is clear and oriented toward practical applications, indicating a substantial contribution to the field.
- Materials and Methods:
Overall, the methodology is solidly structured, emphasizing scientific rigor and thoroughness, making it suitable for replication in future research. That said, there are a few questions that arise:
1. How did you ensure that the selected seeds were representative of the natural variability within each species? Were there any measures taken to account for morphological differences?
2. Can you provide details on the specifications of the camera setup used for image capture? What measures were taken to minimize noise or artifacts during image acquisition?
3. Could you explain the rationale behind using the HSV color space for background removal? How did you determine the specific HSV range for the background, and were there any challenges encountered in this step?
4. Can you elaborate on how you approached hyperparameter tuning beyond setting the learning rate and batch size? Were there systematic methods followed, such as grid search or Bayesian optimization?
- Results:
Overall, the section effectively communicates the models' capabilities and limitations, paving the way for potential improvements and future research. That said, there are a few questions that arise:
1. Were there any specific traits (e.g., color, size, shape) in the misclassified seeds indicated in the confusion matrices that might guide future model adjustments?
2. Given the observed overlap in the t-SNE distributions, what strategies do you propose to improve the models' ability to distinguish between visually similar species in broader datasets?
3. The AUC values were very high across models. Can you discuss how these values translate into meaningful insights regarding model performance, particularly in a practical context?
- Discussion:
This section effectively synthesizes the study's findings, providing insights into the performance of the CNN models in classifying RGB images of Ziziphus seed species. However, the authors wisely acknowledge the limitations of their study, such as the relatively narrow focus on three models and the need for further feature analysis. This sets a constructive tone for future research directions, including enhancing model robustness and exploring new architectures. That said, there are a few questions that arise:
1. You mentioned the limited variety of models used in this study. What specific characteristics or architectures do you envision exploring in future research to enhance classification performance further?
2. How do you envision integrating the developed model into a portable device? What technical specifications or considerations are necessary to ensure the model performs reliably in a field setting?
3. Apart from exploring advanced architectures, do you plan to incorporate other modalities (e.g., infrared, multispectral) for classification to enhance the robustness of the models? If so, what are some potential advantages and challenges of this approach?
4. Are there plans to conduct an analysis of feature importance to better understand which characteristics (e.g., color, texture) the models relied upon most for classification, particularly for Z. jujuba Mill.?
- Conclusions:
1. What strategies do you suggest to implement to test the robustness of the models in real-world settings, especially with variations in lighting conditions, camera quality, or seed condition?
2. In your future research, you mention the potential to classify a wider array of medicinal materials. What specific types of materials do you intend to explore, and what are the anticipated challenges? Please mention more details.
3. As the applications of this research could impact the management and certification of medicinal materials, how do you plan to address any ethical considerations or implications arising from its implementation?
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Comments and Suggestions for AuthorsThis manuscript introduces a classification model based on deep learning technology to distinguish RGB images of seeds from Ziziphus jujuba Mill. var. spinosa, Ziziphus mauritiana Lam., and Hovenia dulcis Thunb.
1. The literature review part is not sufficient to further explain the current research status of deep learning classification models, especially the research on large models.
2. The paper is not innovative. Only three methods are used in the experiment for comparison, and the results are not very different, which is not convincing.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
Comments and Suggestions for Authors
The topic of the study is promising; however, it requires significant modifications. The authors need to revise specific sections with greater attention.
- Lines (15-17): Please describe the outcomes for all the proposed models, highlighting how these results demonstrate the novelty and unique contributions of this study.
- Add additional keywords to comprehensively address the manuscript's objectives.
- The main and submain objectives should be clearer for readers.
- It is unnecessary to mention the graphical abstract when outlining the objectives of this work. Simply upload the graphical abstract to be included in the journal system.
- Lines (99-116) would be better presented as a single paragraph.
- If possible, please add citations to lines (138-152).
- Please include details on how to enhance images using data augmentation and specify the data augmentation protocol followed by the authors.
- Kindly specify the names of the layers within Figure 5.
- Equations (7 and 8) are not essential and should be removed.
- “Model performance evaluation”, this section should be shortened, as the current format is difficult to understand.
- The title "3.1. Model Training" can be adjusted to a more appropriate expression.
- Model losses should be incorporated into Table 2.
- Figures 7 and 9 should be zoomed in to make all the details clearer.
- The discussion lacks sufficient citations; please support the current outcomes and compare them with previous studies. Kindly address and resolve this issue.
- Revise the conclusion to avoid making it similar to the abstract.
This paper requires significant revisions before it can be considered complete. Therefore, I recommend a major revision. Thank you.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 4 Report
Comments and Suggestions for AuthorsThe study investigated ‘Deep Learning-Based Model for Effective Classification of 2 Ziziphus jujuba Mill. Using RGB Images’ which was interesting, and good results are being reported. However, the following issues must be resolved
1. Full stop should be removed from after ‘Mill’ in the title
2. Line 17, avoid ‘we’ in the abstract
3. Please, maintain a 200 word in the abstract
4. Line 20 and 21 …. These results demonstrate that DenseNet-121 is the most practical model for the verification of medicinal plants ……. Authors could say is the best model among the models used in this study. There may be other models which may be better
5. Line 21, avoid the use of ‘our’ please state the plant
6. Line 30 and 31; The use of medicinal plants remains active today; in 2010 ….. please, recast the sentence appropriately and remove the ;
7. Lines 63 – 65; Recent advancements in computer vision and deep learning technologies have led to the development of models that can quickly and cost-effectively identify targets without destruction ……… This is a statement of fact, and it requires references. The following references can be considered
https://doi.org/10.32604/cmes.2022.022088
https://doi.org/10.3389/fphy.2021.644450
https://doi.org/10.48550/arXiv.2406.02291
8. Line 111 …… what is Suan Zao Ren ?
9. Line 159; what are the model's input data ? what percentage was used for training, validation and testing, respectively ?
10. I expect that the model evaluation to be more thoroughly done…. Like what’s the RMSE, MBE, NRMSE and R-square. These are not well considered
11. Line 240 ….. 3.1. Model training ………… This sub-section topic should be more elaborate than this, improve it
12. This is applicable till section 3.5
Comments on the Quality of English Language
Moderate Editing is Required
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe revised manuscript and the authors' responses have been reviewed, and the implemented changes have been thoroughly examined. I am pleased to confirm that the article is suitable for acceptance in its current form.
Reviewer 3 Report
Comments and Suggestions for AuthorsThe authors improved the manuscript according to my comments. It can be accepted for publication.
Reviewer 4 Report
Comments and Suggestions for AuthorsNo further issues
