SpiderID_APP: A User-Friendly APP for Spider Identification in Taiwan Using YOLO-Based Deep Learning Models

Round 1
Reviewer 1 Report
Comments and Suggestions for Author(WSC, World40 Spider Catalog, https://wsc.nmbe.ch/, accessed on July 17, 2022) should be listed only in the reference list as you have already specified [3]. Several references are not developed and thus they do not provide enough info. E.g., ‘While image-based recognition approach has demonstrated potential [10].’ Check language issues. E.g., ‘stem from the inherent challenges of image-based taxonomic classification but also potential human error’ – add ‘from’ before ‘potential’. And punctuation issues. E.g., ‘suggests that with the right refinements, image-based tools’ – add comma after ‘that’. ‘it's crucial’, ‘it can't yet’, ‘it's pivotal to underscore that it doesn't’ – avoid word abbreviations. “the finer microscopic characteristics” – page for the quote is needed. The in-text citation system is variable, either with name + number or with number. E.g., ‘compared to other fields [23,24]. It is observed that accuracy evaluation on testing datasets in such studies is high due to the low diversity in the dataset used. For example, Sinnott et al. 2020 [24] used’. The figures should be improved, unified as style, and thoroughly explained. More development and depth of the methodology and analysis are needed. ‘Data collection was further enchanted’ – you meant ‘enhanced’, right? ‘such as one provided by iNaturalist’ – the one. ‘developed in a Python environment [63], ,’ – remove the second comma. ‘It simplifies, but the depth of taxonomy’ – simplifies what? Check for no or extra spaces throughout the manuscript. E.g., ‘high-level features [68] ,’ ‘application [23] , while’. ‘It has been suggested in several deep-learning studies that an imbalance in the dataset can lead to a decrease in the overall performance of the model [75].’ – ‘studies’, but you cite only one. ‘Data balancing strategies such as over-sampling [76] and under-sampling [76]’ – why citing the same source twice in a row? The conclusion, too short, should clarify the main contribution of the paper and the value added to the field. The reference list is not properly edited, with missing info.
Comments on the Quality of English LanguageCheck language issues. E.g., ‘stem from the inherent challenges of image-based taxonomic classification but also potential human error’ – add ‘from’ before ‘potential’. ‘it's crucial’, ‘it can't yet’, ‘it's pivotal to underscore that it doesn't’ – avoid word abbreviations. ‘Data collection was further enchanted’ – you meant ‘enhanced’, right? ‘such as one provided by iNaturalist’ – the one. ‘It simplifies, but the depth of taxonomy’ – simplifies what?

Author Response
Comments and Suggestions for Authors
The authors thanked the reviewer for your detailed comments and constructive suggestions. Therefore, the authors made modifications to the manuscript according to the comments and suggestions, particularly in the following positions:
The figures should be improved, unified as style, and thoroughly explained.
Thank you for the suggestions. Most of the figures in the manuscript, including Figure 3 to Figure 8 were changed to improve the quality and unify the style, such as the color palettes and caption, of each figure with other figures as the reviewer suggested. In addition, thorough explanations were also added to the figures to provide more information to the readers regarding the contents of the figures.
(WSC, World40 Spider Catalog, https://wsc.nmbe.ch/, accessed on July 17, 2022) should be listed only in the reference list as you have already specified [3].
The authors appreciated the comment. Initially, the information regarding the URL and accessed date were provided to give more information regarding the reference, which is the World Spider Catalog. However, the information in line 40 was removed from the text since it is already specified in the References section as the reviewer suggested.
Several references are not developed and thus they do not provide enough info. E.g., ‘While image-based recognition approach has demonstrated potential [10].’
Thank you for the correction. It is true that the information provided in the manuscript regarding this issue was too scarce, which might confuse some readers. Therefore, in the updated manuscript, specifically in lines 59-60, the authors added a brief explanation regarding the references to clarify the potencies of the demonstrated image-based recognition approach.
Check language issues. E.g., ‘stem from the inherent challenges of image-based taxonomic classification but also potential human error’ – add ‘from’ before ‘potential’.
The authors thanked the reviewer for the correction regarding the language. Thus, the word ‘from’ was added before ‘potential’ in the sentence in line 65 as the reviewer suggested.
And punctuation issues. E.g., ‘suggests that with the right refinements, image-based tools’ – add comma after ‘that’.
Thank you for another correction regarding the punctuation of the sentence. A comma was added after ‘that’ in the sentence from line 72 according to the reviewer’s suggestion.
‘it’s crucial’, ‘it can’t yet’, ‘it’s pivotal to underscore that it doesn’t’ – avoid word abbreviations.
The authors agreed that word abbreviations are not appropriate to be used in the formal research article. Therefore, the word abbreviations in lines 99, 110, 561, and 689 were revised to avoid the abbreviations.
“the finer microscopic characteristics” – page for the quote is needed.
Thank you for the reminder. The original quote and its translation were added to the text in Lines 105-108 to provide more details regarding the term “the finer microscopic characteristics’ mentioned in the text. In addition, the page number where the quote is written in the book was also added to the citation.
The in-text citation system is variable, either with name + number or with number. E.g., ‘compared to other fields [23,24]. It is observed that accuracy evaluation on testing datasets in such studies is high due to the low diversity in the dataset used. For example, Sinnott et al. 2020 [24] used’.
The authors appreciated the detailed check regarding the citations in the manuscript. The authors admitted that some mistakes were made in terms of the in-text citation which caused inconsistencies in its format as mentioned by the reviewer. Therefore, corrections in some parts of the text were made. For example, the mentioned in-text citation in lines 118-119 has now become “Sinnott et al. in 2020 has used…”.
‘Data collection was further enchanted’ – you meant ‘enhanced’, right?
Thank you for pointing out this mistyping. As the reviewer mentioned, the correct word should be ‘enhanced’ instead of ‘enchanted’. Therefore, a correction was made regarding this mistype in line 157.
‘such as one provided by iNaturalist’ – the one. ‘developed in a Python environment [63], ,’ – remove the second comma. Check for no or extra spaces throughout the manuscript. E.g., ‘high-level features [68] ,’ ‘application [23] , while’.
The authors appreciated the reviewer for thoroughly checking the writing in the whole manuscript. The authors had tried their best to double-check the manuscript and made some corrections in lines 157, 550, 580, and 685, regarding the writing, including removing the extra spaces and second comma as the reviewer pointed out.
‘It simplifies, but the depth of taxonomy’ – simplifies what?
Thank you for the question. The authors admitted that the sentence in line 569-573 was not clear. Therefore, revisions were done to clarify the idea of the current method that could simplify the process of initial screening, however, the depth of taxonomy lies beyond its scope.
‘It has been suggested in several deep-learning studies that an imbalance in the dataset can lead to a decrease in the overall performance of the model [75].’ – ‘studies’, but you cite only one.
The authors appreciated the comment and realized that a reference was not cited in the sentence in lines 603-604. Thus, the intended reference, which was another systematic review (Japkowicz, N.; Stephen, S. The class imbalance problem: A systematic study. Intelligent data analysis 2002, 6, 429-449.) was added to the manuscript to further support the statement.
‘Data balancing strategies such as over-sampling [76] and under-sampling [76]’ – why citing the same source twice in a row?
Thank you for the reminder. The authors did not intend to cite the same source twice in a row in the sentence in line 612. Therefore, according to the reviewer’s suggestion, the first citation was removed from the sentence.
The conclusion, too short, should clarify the main contribution of the paper and the value added to the field.
The authors appreciated the critique and suggestions. Thus, in the updated manuscript, the authors added some sentences to clarify the main contribution and the value added to the field of the current paper, which is introducing an innovative tool that combines technological advancement with biological research to the field of arachnology in lines 742-764.
The reference list is not properly edited, with missing info.
Thank you for the comment. The authors thoroughly checked the whole references section (lines 771-959) and modified the reference list to fill in the missing information regarding the references, such as the page numbers, and unify the citation styles of the references.

 Reviewer 2 Report
Comments and Suggestions for AuthorsIn the manuscript “SpiderID_APP: An user-friendly APP for spider identification in Taiwan using YOLO-based deep learning models” the authors applied a deep learning-based approach using the YOLOv7 framework to provide an efficient and user-friendly identification tool for spider species found in Taiwan called Spider Identification APP (SpiderID_APP). This manuscript is well organized, and the drawn conclusions are coherent with the obtained results. The paper was well written!
 Lines 33 – 34: Please, arrange the keywords alphabetically.
Lines 76 – 78: I think that you should these important references as examples to support your sentence: “Modern biological research's breadth is vast, but gathering large-scale biological data and long-term monitoring often hits resource constraints in which citizen science presents a compelling solution”. I would like to suggest:
Di Febbraro, M., et al., (2023). Different facets of the same niche: Integrating citizen science and scientific survey data to predict biological invasion risk under multiple global change drivers. Global Change Biology, 29(19), 5509-5523.
Nanglu, K., et al., (2023). The nature of science: The fundamental role of natural history in ecology, evolution, conservation, and education. Ecology and Evolution, 13(10), e10621.
Lines 97 – 99: I think that you should these important references as examples to support your sentence: “Using deep learning model training on a large dataset of labeled images, which can identify the key features and provide the probability of detection based on such extracted features, is essential for the task.”. I would like to suggest:
Caci, G., et al., (2013). Spotting the right spot: computer-aided individual identification of the threatened cerambycid beetle Rosalia alpina. Journal of insect conservation, 17, 787-795.
Willi, M., et al., (2019). Identifying animal species in camera trap images using deep learning and citizen science. Methods in Ecology and Evolution, 10(1), 80-91.
Lines 123 – 125: Please, explain in detail you hypothesis and predictions.
Lines 136 – 511: Well written!
Lines 513 – 669: I think that this part of the manuscript should be expanded to include a discussion also with other methodology of identification based on tools/packages/software. 
Comments on the Quality of English LanguageThis manuscript is well written!

Author Response
Comments and Suggestions for Authors
In the manuscript “SpiderID_APP: An user-friendly APP for spider identification in Taiwan using YOLO-based deep learning models” the authors applied a deep learning-based approach using the YOLOv7 framework to provide an efficient and user-friendly identification tool for spider species found in Taiwan called Spider Identification APP (SpiderID_APP). This manuscript is well organized, and the drawn conclusions are coherent with the obtained results. The paper was well written!
The authors are deeply grateful for your positive assessment and the encouraging remarks provided. They are particularly pleased to note the acknowledgment of the manuscript's organization and the coherence between the conclusions and the results obtained. The authors took great care in ensuring that the paper was not only informative but also accessible and well-articulated. Such feedback is highly valued and serves as a significant motivation for the team. They extend their heartfelt thanks to the reviewer for their time and effort in reviewing the manuscript and for recognizing the efforts put into developing the Spider Identification APP (SpiderID_APP).
Lines 33 – 34: Please, arrange the keywords alphabetically.
Thank you for the correction. The keywords listed in lines 33-34 were arranged alphabetically in the updated manuscript, which now are ordered as biodiversity, deep learning, genera identification, iNaturalist, spider identification, taxonomy identification, YOLOv7.
Lines 76 – 78: I think that you should these important references as examples to support your sentence: “Modern biological research’s breadth is vast, but gathering large-scale biological data and long-term monitoring often hits resource constraints in which citizen science presents a compelling solution”. I would like to suggest:
Di Febbraro, M., et al., (2023). Different facets of the same niche: Integrating citizen science and scientific survey data to predict biological invasion risk under multiple global change drivers. Global Change Biology, 29(19), 5509-5523.
Nanglu, K., et al., (2023). The nature of science: The fundamental role of natural history in ecology, evolution, conservation, and education. Ecology and Evolution, 13(10), e10621.
The authors appreciated the constructive suggestions from the reviewers and thanked them for pointing out some important references as examples to support the sentence in lines 76-68. After thoroughly reading and comprehending the suggested references by Febbraro et al. (2023) and Nanglu et al. (2023), the authors decided to include both of the references to the manuscript since they discuss the role of citizen science in addressing resource constraints in large-scale biological research, highlighting the importance of these studies in underscoring the vital role of citizen science in various fields of biological research, and thus, are instrumental in elevating the depth and quality of the current work.
Lines 97 – 99: I think that you should these important references as examples to support your sentence: “Using deep learning model training on a large dataset of labeled images, which can identify the key features and provide the probability of detection based on such extracted features, is essential for the task.”. I would like to suggest:
Caci, G., et al., (2013). Spotting the right spot: computer-aided individual identification of the threatened cerambycid beetle Rosalia alpina. Journal of insect conservation, 17, 787-795.
Willi, M., et al., (2019). Identifying animal species in camera trap images using deep learning and citizen science. Methods in Ecology and Evolution, 10(1), 80-91.
Thank you again for providing insightful suggestions to include additional necessary references in the manuscript to support the mentioned sentence in lines 97-99. The authors had read both of the studies reported by Caci et al. (2013) and Willi et al. (2019) acknowledged the relevance of these studies and recognized how they can substantiate the statement. Thus, the authors concluded and considered the importance of these references to be included in the manuscript due to the results that clearly demonstrated the practical application and effectiveness of deep learning models in species identification, which directly supports the point being made in the manuscript, significantly aiding in enhancing the manuscript's academic rigor and contextual depth.
Lines 123 – 125: Please, explain in detail you hypothesis and predictions.
The authors appreciated the suggestions to explain the hypothesis and predictions of the current study. Based on the suggestion, the authors wrote a new paragraph in the introduction section that elaborates on the foundational hypothesis behind the present study and clearly outlines the specific predictions they sought to test. The hypothesis in this study was by applying YOLO models, a deep learning-based framework for spider classification using image datasets available on online open-source databases could be conducted and predicted that the method is effective for conducting a genus-level spider identification. The authors hoped that this section now provides a comprehensive understanding of the underlying assumptions and expected outcomes of their research and this amendment could greatly enhance the clarity and depth of the manuscript.
Lines 513 – 669: I think that this part of the manuscript should be expanded to include a discussion also with other methodology of identification based on tools/packages/software.
Thank you for the suggestion. The authors strongly agree with the reviewer that a discussion is required to be added in the mentioned part of the manuscript to include a discussion with other methodologies of identification that were based on other tools/packages/software. Therefore, the discussion section was rewritten to provide more information regarding the previous academic and non-academic spider identification methodologies and applications (Lines 684-694). This information is necessary to provide a comprehensive overview of the current landscape in spider identification technology drawing comparisons and highlighting the unique aspects of their approach in the context of existing methods. The authors believed that these additions could enhance the discussion section, offering the readers a broader perspective on the topic.

Reviewer 3 Report
Comments and Suggestions for AuthorsAbstract Evaluation:
The manuscript provides a clear paper overview, emphasizing the importance of accurate and rapid spider taxonomy identification. The use of deep learning, particularly the YOLOv7 framework, for developing the Spider Identification APP is a relevant and modern approach to tackling the challenges in spider species identification. It is necessary to review and improve some aspects detailed below.
Strengths:
§  Clear Problem Statement: The abstract effectively communicates the challenge of spider identification due to morphological similarities, highlighting the need for advanced taxonomic methods.
§  Innovative Approach: The use of YOLOv7 and training on a substantial dataset from iNaturalist demonstrates a modern and innovative approach to spider species identification, making it a valuable contribution to the field.
§  Practical Utility: The emphasis on creating an efficient and user-friendly identification tool (SpiderID_APP) is commendable, catering to both researchers and citizen scientists, potentially broadening the scope of spider identification efforts in Taiwan.
§  Reference to iNaturalist: The use of 24,000 images from the iNaturalist database for training adds credibility to the study, ensuring a diverse and representative dataset for model development.
Points for Improvement:
1.      Better visualization of figures 7 and 8: it is necessary to improve the size of the letters on the axes and the size of the images for a better visualization of the data.
2.      Lack of Technical Details: The abstract lacks specific technical details regarding the architecture modifications to YOLOv7 or the evaluation metrics used. Including such information would enhance the understanding of the methodology and the robustness of the model.
3.      Validation and Generalization: While the paper mentions the accuracy of the SpiderID_APP being on par with iNaturalist, it would be beneficial to include details about the validation process and the model's generalization capabilities to different spider populations.
4.      Data Sources and Representation: The abstract briefly mentions the use of 120 generated classifications for Taiwan spider species but lacks information on the representativeness of these classifications and how well they cover the diversity of spider species in Taiwan.
5.      Discussion of Limitations: The manuscript briefly touches upon the necessity of additional methods for detailed species identification but could benefit from a more detailed discussion of the limitations and potential challenges faced by the SpiderID_APP.

Author Response
Comments and Suggestions for Authors
The manuscript provides a clear paper overview, emphasizing the importance of accurate and rapid spider taxonomy identification. The use of deep learning, particularly the YOLOv7 framework, for developing the Spider Identification APP is a relevant and modern approach to tackling the challenges in spider species identification. It is necessary to review and improve some aspects detailed below.
Strengths:

1. Clear Problem Statement: The abstract effectively communicates the challenge of spider identification due to morphological similarities, highlighting the need for advanced taxonomic methods.

2. Innovative Approach: The use of YOLOv7 and training on a substantial dataset from iNaturalist demonstrates a modern and innovative approach to spider species identification, making it a valuable contribution to the field.

3. Practical Utility: The emphasis on creating an efficient and user-friendly identification tool (SpiderID_APP) is commendable, catering to both researchers and citizen scientists, potentially broadening the scope of spider identification efforts in Taiwan.

4. Reference to iNaturalist: The use of 24,000 images from the iNaturalist database for training adds credibility to the study, ensuring a diverse and representative dataset for model development.

The authors express their sincere gratitude for your encouraging and detailed feedback on the manuscript. We are particularly appreciative of your recognition of the clear problem statement in the abstract, which was crafted to effectively communicate the challenges in spider identification due to their morphological similarities. Your acknowledgment of the innovative approach taken by using the YOLOv7 framework and training on a substantial dataset from iNaturalist is highly motivating. The authors aimed to bring a modern and innovative solution to the field of spider species identification, and your recognition of this effort is greatly valued.
We are also grateful for your appreciation of the practical utility of the SpiderID_APP. The primary goal was to create an efficient and user-friendly tool that caters not only to researchers but also to citizen scientists. This feedback confirms that the intended purpose of broadening the scope of spider identification efforts in Taiwan is being realized.
Lastly, your mention of the use of 24,000 images from the iNaturalist database as a strength adds further credibility to our study. The authors were meticulous in ensuring a diverse and representative dataset for model development, and your recognition of this effort is very much appreciated.
The authors thank you for your insightful and constructive comments, which are invaluable in guiding further improvements to the manuscript.
Points for Improvement:

1. Better visualization of figures 7 and 8: it is necessary to improve the size of the letters on the axes and the size of the images for a better visualization of the data.

The authors appreciated the suggestion regarding the enhancement of Figures 7 and 8 and were fully aware of the necessity of improving the size of the letters on the axes and the size of the images for better visualization of the data. Therefore, the sizes of the axes and images in the mentioned figures were carefully adjusted to enhance the clarity and readability of the data. The authors hope that the changes are appropriate and sufficient to help the readers understand the data presented in the figures.

2. Lack of Technical Details: The abstract lacks specific technical details regarding the architecture modifications to YOLOv7 or the evaluation metrics used. Including such information would enhance the understanding of the methodology and the robustness of the model.

Thank you for the reminder. The authors understood the reviewer’s point of view in adding some specific technical details in the abstract, particularly about the modifications to the YOLOv7 architecture and the evaluation metrics employed. Although the authors also considered the importance of this matter, one has to keep in mind that the current work encompassed a comprehensive process that involved not only gathering and labeling a substantial image dataset but also ensuring the quality of these images for effective training, which extended to employing the base architectures of YOLOv5, YOLOv6, YOLOv7, and YOLOv8, followed by a meticulous comparison and optimization of various training parameters. Therefore, the authors believed that it would be more appropriate to briefly mention some of these crucial technical aspects in the abstract and the authors felt that these changes are sufficient enough to provide a clearer understanding of the methodology and reinforce the robustness of the model in a short explanation. However, the details of this matter are extensively discussed in the body of the paper to enhance the understanding of the methodology and the robustness of the model.

3. Validation and Generalization: While the paper mentions the accuracy of the SpiderID_APP being on par with iNaturalist, it would be beneficial to include details about the validation process and the model's generalization capabilities to different spider populations.

The authors appreciated the suggestion to include some details about the validation process and the model's generalization capabilities to different spider populations. Therefore, revisions were made to clarify the internal and external validation processes in the discussion section. In addition, “Section 5. Potential Limitations and future work” was also modified to incorporate the view on the model’s generalization capabilities to different spider populations (the details regarding this matter can be found in the response to the comment in point number five).

4. Data Sources and Representation: The abstract briefly mentions the use of 120 generated classifications for Taiwan spider species but lacks information on the representativeness of these classifications and how well they cover the diversity of spider species in Taiwan.

Thank you for pointing out this mistake that was caused by the oversight. Therefore, the authors added some information regarding the representativeness of these classifications and how well they cover the diversity of spider species in Taiwan in the discussion section. The changes include the addition of a detailed discussion about the representativeness of 120 genera within the Taiwan spider species, emphasizing the high frequency of encountering such 120 genera leading to a high percentage of their records within databases.

5. Discussion of Limitations: The manuscript briefly touches upon the necessity of additional methods for detailed species identification but could benefit from a more detailed discussion of the limitations and potential challenges faced by the SpiderID_APP.

The authors appreciated the suggestion. In the previous version of the manuscript, the authors admitted that there was only a brief touch upon the necessity of additional methods for detailed species identification. Thus, revisions were done in the mentioned section to clearly highlight the limitations faced by the current methodology, including the inability to evaluate the models of spider populations from different geographical regions and the future works in dealing with such shortcomings (Lines 709-721). In addition, more comprehensive analyses to address the potential challenges faced by the SpiderID_APP were also added to the manuscript (Lines 730-740).

Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThis revised version can be published.