HCFormer: A Lightweight Pest Detection Model Combining CNN and ViT
Round 1
Reviewer 1 Report (New Reviewer)
Comments and Suggestions for AuthorsDear Authors,
I will like to congratulate you for the brilliant work. Climate change and globalization are helping in the change of pests and pathogens evolutional trends. This can make them become more virulent with much negative impact on crops thus favouring food insecurity. Controlling these pests through conventional pesticide presents a double negative impact on environment and economy (increased crop production cost). Effective pest management is essential for good crop yields, thus accurate pest identification and classification are necessary.
This work presents an interesting approach for the pest detection contributing towards a proper and environmental control strategy. With the advancement in technology, leveraging ML and AI to tackle pest and disease is an appropriate approach. It is an interesting result for HCFormer to show improved accuracy over SENet, CrossViT, and YOLOv8.
For the purpose of scientific publication, I have provided some comments in the paragraphs that follows.
General comments:
1- If I understand correctly, only images from Flickr Database was used for this analysis. Is there a possibility of getting images of the same pest from different database? Will images from different source product the same result with HCFormer.
2- I propose you carefully check the sentence structures (especially punctuation marks).
3- It will be a good idea to check the completeness of the information provided in figures and table headings.
4- How easy can this method be applicable for local farmers? This worth discussing.
5- I strongly recommend: if possible, it will be a good idea to train HCFormer using picture directly taken from the field side-by-side that from database.
6- I strongly recommend this: if possible, it will be a good idea to train the model-using picture directly taken from the field side-by-side that from database.
Specific comments:
1- Line 5: (college of Artificial Intelligence), changing “college” to College.
2- Line 23: …outperforming mainstream detection models overall. May be better to write it as “…outperforming the overall mainstream detection models.” I hope this do not change the sense.
3- Line 35: …global economy USD 220 billion and USD 70 billion, respectively. Will it be nice to write it as “…global economy $220 and $70 billion, respectively.”
4- Line 37-39 and 57-60: Please consider providing an appropriate citation.
5- Line 43: …treatment of different pest types are essential for effective pest management. Consider writing, as “…treatment of different types of pest are essential for effective pest management.”
6- Line 45: …for non-professionals, even experienced observers,… can be deleted, this will not change the sense of the sentence.
7- Line 142-145: Please consider mentioning these lines under result section.
8- Line 258-278: This information has been provided under the introduction section. Please consider mentioning precisely what was done under methodology.
9- Line 482- 483: Check the figure description for precise information. This title should let the figure stand-alone. There are a, b, and c, for figure 8 with a lower and an upper panels. I think more information is needed.
10- Line 684: Consider checking reference number 30 for editing. There is an image of a rocket inserted…I do not know if the is allow for referencing GitHub?
Comments on the Quality of English LanguageI strongly recommend checking the sentence structure (especially punctuation marks)
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report (New Reviewer)
Comments and Suggestions for AuthorsThis paper presents a very interesting study that proposes the use of a novel lightweight pest detection model that combines ViT and CNN. The manuscript is generally well-structured and well-written. The importance, novelty and overall quality of the paper are adequate. The experiments seem to have been conducted correctly using the right methodology. However, more analysis could have been conducted to confirm the efficiency of the model. Moreover, discussion and the use of bibliography are relatively poor. There are some points that need to be added, corrected or improved. Please find below the comments and suggestions that need to be addressed:
Lines 11-28. The abstract is a little larger than it should be but is well structured and comprehensive. Therefore, it is fine even if it exceeds 200 words. You might substitute the phrase “together with” by “along with” in line 11.
Line 34. The term “insect assaults”, although is not completely wrong, is not commonly used and is not proper in Pest science.
Line 48. Citation/citations needed at the end of the sentence.
Lines 253-254. There is a mistake in Figure 2. There is no image (d) and image (e). Instead, there are two images displaying the letter (d). Also, you should probably switch the position of the two images of the bottom line as the one on the left seems to be “locally zoomed”.
Lines 284, 293, 325, 375, 412, 462, 471, 496, 536 etc. It is better to avoid sentences with personal reference to the authors (“we” is mentioned too many times in the text). Passive voice is preferred.
Lines 505-507. In figure 9 the image quality is poor. Also, it would be helpful to provide the species’ Latin names. The legend could also include more information.
Lines 514-532. Yes indeed, the HCFormer displayed excellent results. But some other models also demonstrated very good results. Did you consider conducting statistical analysis to compare results?
Lines 542-552. Although this is a short discussion, a larger - separate discussion section could be added to discus the highlights of your findings and make comparisons with other studies/models.
Lines 552-558. More information and discussion about that should be provided.
Line 571. You mention “strong practicality”. More information about how the model can be practically applied in agroecosystems should be provided.
References. It is understandable that as a specialised topic there aren’t thousands of sources available. However, the number of references is limited. More studies could be referred for sure. On the top of that, many of the existing references refer to conference proceedings or even websites. I recommend that you add more citations of research articles where possible.
Lines 684-685. The reference no. 30 is presented in a form which is not acceptable.
Comments for author File: Comments.pdf
The use of the English language is generally good throughout the manuscript. However, there are a few points that should be revised to ensure more proper terminology.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report (New Reviewer)
Comments and Suggestions for AuthorsThe manuscript discusses the employment of a lightweight pest detection network, HCFormer, combining convolutional neural networks (CNNs) and a vision transformer (ViT) in data preprocessing to reduce computational latency. However, there are major comments that could enhance the manuscript.
1) There is more data in this area and references that could be added to enrich the manuscript.
2) A clear flowchart for the methodology is required to understand the process that was created to reach the developed approach.
3) Is the model validated through several images existing in the literature review to maintain the validity?
4) To what extent does the resolution of the image influence the model
5) Is it applicable to structure detection of cracks or corrosion if possible
6) Can the model measure or detect the size of the pest within the image as exactly in the actual life
7) The author should explain more about how can the model detect several pests in the same image and if it is possible to have different types of pests in the same image
8) What are the limitations of this approach or this study
9) The results and discussion should be deeper
10) The amended conclusion should be updated according to the modification.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report (New Reviewer)
Comments and Suggestions for Authorsno further comments
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThere are still too many issues to be revised in this manuscript.
1. English in this manuscript should be improved. There are too many errors.
2. In the Abstract, ‘By epoch 100, the loss 20 approaches zero, indicating that our model quickly converges. ‘ need be corrected.
3. How did YOLOv5 and YOLOv8 work?
4. The format of ’Malek, M.A. et al.[7]’ should be changed to ‘Malek et al.[7], ’. Other references should be revised too.
5. References [21] and [22] did not focus on Pest Classification.
6. As shown in Table 1, the numbers of the original images is 2760. This should be mentioned in the text.
7. ‘We divided the dataset as shown in Table 3,’ but what does Table 3 show?
8. ‘To assess training performance, 347 we have displayed the accuracy and the loss curve over the epochs in the figure below: ’, which figure?
9. Although the References were checked, there are still many errors in the References section, such as Reference [6], [18], [29]….
Comments on the Quality of English LanguageEnglish in this manuscript should be improved. There are too many errors.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThere is little scientific justification for the classification of twelve selected very different types of pests using a dataset from Kaggle. High accuracy can be expected. Originality and significance of content are very low.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
