All reviews of published articles are made public. This includes manuscript files, peer review comments, author rebuttals and revised materials. Note: This was optional for articles submitted before 13 February 2023.
Peer reviewers are encouraged (but not required) to provide their names to the authors when submitting their peer review. If they agree to provide their name, then their personal profile page will reflect a public acknowledgment that they performed a review (even if the article is rejected). If the article is accepted, then reviewers who provided their name will be associated with the article itself.
The manuscript read well, and the authors have added additional supporting evidence in the Results and Discussion to support their findings. Both reviewers had similar concerns, and Reviewer 1 confirmed they are happy with the revisions. Reviewer 2 did not re-review this version, but I have checked the revisions, and can confirm that the authors have addressed Reviewer 2's comments. This manuscript is ready for publication.
[# PeerJ Staff Note - this decision was reviewed and approved by Mehmet Cunkas, a PeerJ Section Editor covering this Section #]
the authors have addressed my comments. I think, this manuscript is now acceptable
Design of the experiments are well elaborated and replicable imho.
Authors worked toward establishing validity of the study and I am satisfied with the steps taken.
None
**PeerJ Staff Note:** Please ensure that all review, editorial, and staff comments are addressed in a response letter and that any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate.
**PeerJ Staff Note:** It is PeerJ policy that additional references suggested during the peer-review process should only be included if the authors agree that they are relevant and useful.
Basic reporting is ok now.
In data collection, authors mentioned that "For each turtle, 20 photographs were taken, focusing on specific characteristics to identify 30 turtles, including nasal stripes (face), position and shape of the infraorbital stripe (side face), and plastron patterns.". The Information, in the form of a table, is required about each parameter for the photograph. For example, angle, direction, etc.
Furthermore, authors would have taken images of each side of the turtle (photograph parameters) twice or more times by suggesting just the guidelines to multiple humans. In this way, the turtles could be identified by multiple different pictures taken by different humans would have been validated. Is it still possible to collect pictures of some of the earlier captured turtles?
Validity is not significant. Please see my previous comment. The validity should be improved.
- The authors have undertaken an important study with the goal of developing a deep learning-based automatic individual identification system for Malayemys khoratensis. This is a meaningful and timely topic with clear potential to advance research in the field. Nevertheless, despite the revisions made, the current manuscript still does not sufficiently support the authors' claims. I encourage the authors to take enough time for substantial revision and refinement. That said, I believe the study has considerable promise. With more time for substantial revision and further refinement, this study could develop into a stronger and more valuable paper. The following comments are provided in the spirit of constructive comments, with the hope that they may assist the authors in more clearly conveying their research objectives and strengthening the overall quality of the study.
- In response to concerns regarding the need for additional data, the authors repeatedly stated that the “real situation is 4–20 images.” However, the amount of data required for effective model training and the number of images typically collected in real situations should be considered separate issues. With respect to the training dataset, although the dataset was augmented with an additional 30 images, the overall quantity remains insufficient due to the fundamental limitation of having too few original images. In real-world applications, the concern may lie less with the quantity of data and more with its quality. In particular, if all training data were collected against a white background, the model may face considerable difficulties when applied in field conditions. Such limitations should be clearly acknowledged and carefully considered, as they may substantially affect the model’s generalizability and practical utility.
- The authors stated during the revision process that they “weight decay for prevention overfit and monitor loss value each epoch.” However, if weight decay was applied to prevent overfitting, this should have been explicitly described in the Materials and Methods section. Moreover, merely reporting that the loss values were monitored is insufficient; the corresponding results should also be provided rather than limiting the explanation to observation. This is particularly important in a study reporting such high performance, as demonstrating the absence of overfitting is critical to substantiating the validity of the results.
- The authors presented evidence of novel characteristic patterns that could be used to identify M. khoratensis, such as facial patterns, suborbital stripes, and carapace patterns, through the confusion matrices in Figures 6–8. However, more concrete evidence is needed to demonstrate that the deep learning model successfully captures and utilizes these novel features. This can be achieved by visually highlighting the specific regions on which the model focuses during detection and classification [1,2]. A representative approach would be the application of heat map visualization techniques, such as Grad-CAM or related variations. Employing such methods to verify alignment between the authors' proposed novel patterns and the model decision-making process would significantly enhance the credibility of the results.
[1] Jiang, H., Zhao, J., Ma, F., Yang, Y., & Yi, R. (2025). Mobile-YOLO: A Lightweight Object Detection Algorithm for Four Categories of Aquatic Organisms. Fishes, 10(7), 348.
[2] Ariyametkul, A., & Paing, M. P. (2025). Analyzing explainability of YOLO-based breast cancer detection using heat map visualizations. Quantitative Imaging in Medicine and Surgery, 15(7), 6252.
(Although reference [2] does not focus on biological subjects, it employs a variety of heat map visualization techniques, and the interpretations provided therein may offer valuable insights for the present study.)
**PeerJ Staff Note:** It is PeerJ policy that additional references suggested during the peer-review process should only be included if the authors are in agreement that they are relevant and useful.
The reviewers have provided substantial insight aimed at improving the submission.
**PeerJ Staff Note:** Please ensure that all review and editorial comments are addressed in a response letter and that any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate.
**Language Note:** The review process has identified that the English language must be improved. PeerJ can provide language editing services - please contact us at [email protected] for pricing (be sure to provide your manuscript number and title). Alternatively, you should make your own arrangements to improve the language quality and provide details in your response letter. – PeerJ Staff
• The manuscript presents an application of YOLOv8 for non-invasive turtle identification, with high accuracy and ecological relevance. The article has multiple grammatical issues which make the manuscript incomprehensible. Furthermore, there are multiple reproducibility concerns, dataset limitations, and non-availability of comparative analysis that are weakening the work. The issues are:
• There are multiple grammatical errors which need correction. Phrasing issues (e.g., thelong in abstract to “the long”, "purposed model" → "proposed model," "genius Malayemys" → "genus Malayemys"). A thorough proofread is recommended.
• The end sentence of methods in abstract “… patterns and enhancing by directional weight parameter to detect individual and recognition accurately” is not clear and looks inaccurate sentence.
• Multiple sentences don’t make sense. For example, “Among the purposed model, a novel set of plastron patterns achieved the highest accuracy, reaching up to 99% individual precision when training (80%) and testing (20%) with 30 Khorat snail-eating turtles.” is incomprehensible
• Sentence on line 45-49 need citations. The following sentence can be induced with proper references in the preceding sentence.
• The significance of the research is still not clear. Why is there a need to use ML to identify the turtles? I think a turtle can be easily identified by a human. So, where can such ML based identification help?
• From Table 1, it is not clear which animal research was conducted. I assume for turtles, but use of beak also points to birds
• In the methods section, why 4 – 20 images were taken. For which turtles 4 and for whom 20? Why not uniform and similar features were captured?
• Line 145 – 152. The future work and conclusions are highlighted prematurely in the material and methods section. The section should be moved to a more appropriate place, preferably at the end. Furthermore, following areas also need improvement:
• Reproducibility: While preprocessing steps are outlined, critical details are missing (e.g., augmentation parameters, anchor box configurations, hyperparameters for training YOLOv8).
• Data Preprocessing: The rationale for resizing images to 640px and manual augmentation is unclear. Justify these choices or reference standard practices.
• Ethical Considerations: Although ethical approval is mentioned, details on minimizing stress during turtle handling (e.g., release protocols, photo capture duration) are lacking.
• Line 244-245: A rationale is missing of how the approach ensures precision and rapid identification?
• Generalizability: The small sample size (30 turtles) and lack of cross-validation raise concerns about overfitting. Testing on a larger, geographically diverse dataset would strengthen validity.
• Comparative Analysis: No comparison is made with other models (e.g., YOLOv5, Faster R-CNN) to justify YOLOv8’s superiority.
• Impact Discussion: While novelty is not required per criteria, a brief discussion of how this method advances conservation efforts would add depth.
1. Expand Dataset: Collaborate with broader networks to increase sample size and diversity.
2. Clarify Figure 10: Include visual examples of K15/K16 plastron patterns to illustrate the false-negative issue.
- This study aims to develop the deep learning-based automatic individual identification of Malayemys khoratensis. The developed model might be useful for addressing the time-consuming and labor-intensive process of manual individual identification in the field. However, the current manuscript should be improved for the publication in this journal. More detailed background explaining the need for the study, description of the research methods used, and discussion of the results are needed. Further comments are listed below.
- A deep learning model was developed for the individual identification of M. khoratensis. However, the manuscript currently provides limited background on the ecological or biological importance of M. khoratensis and offers minimal explanation regarding the necessity of individual identification for this species. Addressing these issues and further strengthening the manuscript might be achieved by expanding the Introduction to incorporate additional background on the ecological or biological significance of M. khoratensis and to provide a more detailed rationale for the importance of individual identification. This additional context could help readers better understanding the the study.
- The YOLOv8 model was used for the individual identification of M. khoratensis in this study. The manuscript does not provide the detailed advantages in the model over alternative approaches. Addition of an explanation of the benefits that led to the selection of the YOLOv8 model is needed.
- The authors manually took images of 30 individuals of the species in this study, with only 4 to 20 images per individual. As suggested by the developer of YOLOv8 (https://github.com/ultralytics/ultralytics/issues/2438), 4 to 20 images would not be enough to recognize objects in new images in the learned features, Therefore, it would be advisable to collect a larger number of images per individual, thereby providing a richer dataset for training to improve model performance and reliability.
- The conditions for training the models were not provided in the Materials and Methods part. This is crucial for the performance of the deep learning models. Authors should provide this information, such as the number of epochs, batch size, and size of input images. These conditions are vital for other researchers to reproduce the results of this study.
- The authors evaluated model performance using the F1 Score. However, the YOLO models are typically assessed using the mean Average Precision (mAP), which better captures both detection and classification accuracy. Notably, studies using the YOLOv8 model have commonly used mAP as the standard metric for performance comparison. [1,2,3]. Therefore, if this study uses the F1 Score, it is essential to provide a clear and detailed description of this choice over mAP, explaining how it might offer advantages or additional insights in the context of the current research.
[1] Casas, G. G., Ismail, Z. H., Limeira, M. M. C., da Silva, A. A. L., & Leite, H. G. (2023). Automatic detection and counting of stacked eucalypt timber using the YOLOv8 model. Forests, 14(12), 2369.
[2] Dave, B., Mori, M., Bathani, A., & Goel, P. (2023). Wild animal detection using YOLOv8. Procedia Computer Science, 230, 100-111.
[3] Ma, J., Guo, J., Zheng, X., & Fang, C. (2024). An Improved Bird Detection Method Using Surveillance Videos from Poyang Lake Based on YOLOv8. Animals, 14(23), 3353.
**PeerJ Staff Note:** It is PeerJ policy that additional references suggested during the peer-review process should only be included if the authors are in agreement that they are relevant and useful.
- According to the results, evaluation metrics such as Accuracy, Precision, Recall, and F1-Score reached a perfect 1.0 for 28 out of the 30 individuals, with only two individuals not achieving this score. While these metrics may indeed indicate very high performance, the fact that nearly all the results are perfect raises concerns about potential overfitting. Overfitting can be influenced by various factors, including the relatively small size of the dataset used for training. To address these concerns and provide a more comprehensive evaluation, it would be beneficial to include additional analyses, such as reporting the loss function values (for instance, box, object, and class losses) to clearly demonstrate that the model did not overfit the training data.
- One of the main contributions of this study is the identification of novel characteristic patterns, such as facial patterns, infraorbital stripes, and plastron patterns that can be used to recognize M. khoratensis. However, the manuscript did not provide results confirming the confidence of these findings. It might be beneficial to include a heat map analysis to validate this contribution. This analysis can visually demonstrate the specific regions which the deep learning model focuses on when detecting and classifying individuals, thereby offering concrete evidence that the model successfully captures and utilizes these novel features.
- This model was developed for field applications to identify individuals of M. khoratensis. To achieve the aim, the background of training images may need to be tailored according to the intended field environment. For instance, in a wild setting, the presence of diverse backgrounds, such as various plants and natural structures, could enhance the performance of the model, whereas, in an interior setting, a controlled, uniform background similar to that used in the current work might prove more beneficial in a certain specific circumstance. Consequently, it is recommended that future studies include a detailed discussion of the intended application scenarios to ensure the model is appropriately validated and optimized for its specific field conditions.
All text and materials provided via this peer-review history page are made available under a Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.