Estimating Tree Health Decline Caused by Ips typographus L. from UAS RGB Images Using a Deep One-Stage Object Detection Neural Network^†

Round 1

Reviewer 1 Report

This study investigates the potential of a neural network (YOLOv4-p5) to detect Norway spruce trees attacked by a bark beetle species (Ips typographus) in Finland. For this purpose, it used different Finnish data sets (RGB drone imagery), partly published previously. As a result, the study shows that it is possible to reliably distinguish healthy trees from infested trees with crown degradation, but not from infested trees without crown degradation. This outcome is totally in line with previous knowledge, thus it isn´t actually novel. The advantage over previous study is the relatively large data set and the application of a recent machine learning approach.

The manuscript is very well and clearly written; some minor remarks are listed below. One aspect that is missing in the manuscript is that the background / motivation behind the study results remains unclear. That is, how UAS may support bark beetle management based on the study results? Would an application be sufficient when done once a year, if so preferably in autumn? If “green-attack” can´t be realized, would it still make sense to sanitize the detected trees? How can results from Finnish conditions be transferred to Central European conditions (with different beetle phenology)? …

 

SPECIFIC COMMENTS:

L22         here “Northern Latitudes”, but “Northern latitudes” in L 45

L116       this paragraph (“Diverse …”) is not introduction and should be skipped here

L133       “UAV” is used here and in L 140, but “UAS” is used elsewhere, please use one term consistently if possible

Fig 1       “Lahti” can be skipped in the map, since “Helsinki” isn´t mentioned; “Lahti” is still visible in the grey background

Fig 2       try to make all panels equal in height, so enlarge extent of b) and c) if possible

L141       correct order and letters in the caption

L179       replace “or under” with “or lower”?

L196      explain “GSD” before using that abbreviation

Tab 3     “Quadcopter” (in the rightmost column) can easily be written in full length without abbreviation

L210       sometime references are given as names in addition to the ref ID, but it is not done consistently; it remains unclear when names are added

L216       check ref to Fig. 3a and b, it seems to be reverse

L271       explain “AP” before using that abbreviation

L285       skip dot after “score”

L314       skip “hyphen“ in “Hyper-parameters”

L338       skip explanation for “AP” here, but provide it when “AP” is used at first instance

Tab 7     skip spaces in line 5 and 7 of the table body in the 1^st column

L365       skip space after “bounding”

Fig 6       skip “orange: infested” from the caption as no infested tree is shown here

Fig 6/7  colored frames would be sufficient without naming “healthy” or “dead”, otherwise it seems inconsistent and it partly overlaps (Fig 6a) and is thus difficult to read; lettering should be centered at the bottom of the panels

Fig 7       “(b) model …” instead of “(b) Model …”, “(c) ground …” instead of “(c) Ground …” to be consistent with Fig 6 caption?

L476       explain “ML” before using that abbreviation

L500       correct author names´ spelling

L511       add “a” after is

L512       add “in [14]” after “used”

L531       explain “SVM” before using that abbreviation

L532       “UAV” is used here, but “UAS” is used elsewhere, please use one term consistently if possible

L566       check grammar (“were potential”?)

L568       correct “YOLOve-p5” to “YOLOv4-p5”

L591       sometime references are given as names in addition to the ref ID, but it is not done consistently; it remains unclear when names are added

L613       “single-stage” is used here, but “one-stage” is used elsewhere, please use one term consistently if possible

L614       please add why point 1) is an important contribution, e.g. because that make the result more robust

L623       check grammar (“our dataset”?) and wording (“overridden”)

L630       this sentence is not needed here, as it points too much on Finnish conditions. It is rather important more generally.

L658       add dot after “inspection”

Author Response

First of all, we would like to thank Reviewer 1 for the very valuable comments. Significant issues were identified and we think that these comments helped us to improve the manuscript significantly.

Comment: This study investigates the potential of a neural network (YOLOv4-p5) to detect Norway spruce trees attacked by a bark beetle species (Ips typographus) in Finland. For this purpose, it used different Finnish data sets (RGB drone imagery), partly published previously. As a result, the study shows that it is possible to reliably distinguish healthy trees from infested trees with crown degradation, but not from infested trees without crown degradation. This outcome is totally in line with previous knowledge, thus it isn´t actually novel. The advantage over previous study is the relatively large data set and the application of a recent machine learning approach.

The manuscript is very well and clearly written; some minor remarks are listed below. One aspect that is missing in the manuscript is that the background / motivation behind the study results remains unclear. That is, how UAS may support bark beetle management based on the study results? Would an application be sufficient when done once a year, if so preferably in autumn? If “green-attack” can´t be realized, would it still make sense to sanitize the detected trees? How can results from Finnish conditions be transferred to Central European conditions (with different beetle phenology)? …

Answer: Thank you for your positive comments. And thank you for raising these important questions. We added discussion in Section 4.4 about these topics.  

SPECIFIC COMMENTS:

L22         here “Northern Latitudes”, but “Northern latitudes” in L 45

Answer: We have corrected these to ”northern latitudes” 

L116       this paragraph (“Diverse …”) is not introduction and should be skipped here

Answer: The last paragraph of Introduction was rewritten based on the comments from all three reviewers. We improved the research aim and research questions. We still give some general text about the implementation and results to follow the mdpi-instrructions that state that: “Finally, briefly mention the main aim of the work and highlight the principal conclusions. “

L133       “UAV” is used here and in L 140, but “UAS” is used elsewhere, please use one term consistently if possible

Answer: UAV was replaced by UAS

Fig 1       “Lahti” can be skipped in the map, since “Helsinki” isn´t mentioned; “Lahti” is still visible in the grey background

Answer: Lahti texts were removed.

Fig 2       try to make all panels equal in height, so enlarge extent of b) and c) if possible

Answer: Orthophotos were resized to the same height and into two columns.

L141       correct order and letters in the caption

Answer: Corrected

L179       replace “or under” with “or lower”?

Answer: Corrected

L196      explain “GSD” before using that abbreviation

Answer: Corrected

Tab 3     “Quadcopter” (in the rightmost column) can easily be written in full length without abbreviation

Answer: Corrected

L210       sometime references are given as names in addition to the ref ID, but it is not done consistently; it remains unclear when names are added

Answer: We decided to remove all names from the references. Based on that we edited the text throughout to fit the new presentation.

L216       check ref to Fig. 3a and b, it seems to be reverse

Answer: The text is consistent with the figure order. We replaced 3a with a less confusing figure and improved the explanation of 3b as follows: “Cropped images with black background were then created “ 

L271       explain “AP” before using that abbreviation

Answer: Corrected. We also added below Equation 3 that “AP@0.5 and AP@0.5:0.95 are explained in Section 2.4.” 

L285       skip dot after “score”

Answer: Corrected

L314       skip “hyphen“ in “Hyper-parameters”

Answer: Corrected

L338       skip explanation for “AP” here, but provide it when “AP” is used at first instance

Answer: Corrected. 

Tab 7     skip spaces in line 5 and 7 of the table body in the 1st column

Answer: Corrected. 

L365       skip space after “bounding”

Answer: Corrected. 

Fig 6       skip “orange: infested” from the caption as no infested tree is shown here

Answer: Corrected. 

Fig 6/7  colored frames would be sufficient without naming “healthy” or “dead”, otherwise it seems inconsistent and it partly overlaps (Fig 6a) and is thus difficult to read; lettering should be centered at the bottom of the panels

Answer: Thank you for this comment. We made some changes to Figures 6 and 7. We changed the order of ground truth (now in 6a, 7a) and result images (6b, 7b, c). We also removed the second example from Figure 7. We decided to keep the dead-infested-healthy texts. These are results of the detections by YOLO and it is a common way to present YOLO-results like this. 6a and 7a are manually labeled images and their classes are presented by the colors of frames. We think that differences in presentation support readers to separate ground truth and detections.

Fig 7       “(b) model …” instead of “(b) Model …”, “(c) ground …” instead of “(c) Ground …” to be consistent with Fig 6 caption?

Answer: Corrected according to the suggested

L476       explain “ML” before using that abbreviation

Answer: We removed ML from the title

L500       correct author names´ spelling

Answer: In this new version there are no author names. 

L511       add “a” after is

Answer: We corrected to form “which was an”

L512       add “in [14]” after “used”

Answer: Corrected

L531       explain “SVM” before using that abbreviation

Answer: Corrected

L532       “UAV” is used here, but “UAS” is used elsewhere, please use one term consistently if possible

Answer: Corrected

L566       check grammar (“were potential”?)

Answer: Changed to “were promising”

L568       correct “YOLOve-p5” to “YOLOv4-p5”

Answer: Corrected

L591       sometime references are given as names in addition to the ref ID, but it is not done consistently; it remains unclear when names are added

Answer: We decided to remove all names from the references. Based on that we edited the text throughout to fit the new presentation.

L613       “single-stage” is used here, but “one-stage” is used elsewhere, please use one term consistently if possible

Answer: Corrected to one-sage

L614       please add why point 1) is an important contribution, e.g. because that make the result more robust

Answer: Thank you for this good and important notice. In general, we have improved this Section 4.4 quite comprehensively not only to discuss our results but also their potential further impacts. For this contribution we added an explanation: “This result indicated that the method robustly adopted diverse datasets. “

L623       check grammar (“our dataset”?) and wording (“overridden”)

Answer: “our” removed. “overridden” is correct term for this sentence.

L630       this sentence is not needed here, as it points too much on Finnish conditions. It is rather important more generally.

Answer: The indicated sentence was removed.

L658       add dot after “inspection”

Answer:  Corrected

Reviewer 2 Report

Dear Authors,

I have finished my review on your paper.

Great job! Please consider my comments given in the following.

Best regards,

Rev.

PAPER IN GENERAL:

General: A good paper which fits to the scope of the journal and which can bring improvements to our knowledge.

Approach: Can it be said that your approach differentiates between trees subjected to infestation with bark beetle or would it be a common one to detect the health of the trees? For instance, there may be other causes due to which the trees get dry and eventually die. If your approach would work the same, then I think that the scope of the paper, the title, part of the introduction and so on, need to be adjusted. Then, there is the question of large-scale deployment. How do you see this problem given that (as I understood) the training-validation-testing required a reduction in image size? How about the further effort to collect UAV data at large scale or to adapt large area remote sensing to your findings?

Format: needs improvements, particularly by moving some text between sections, and enhancing the figures throughout the manuscript.

Language: needs improvement in phrasing and English throughout the text. I’ve added specific comments to those lines and sections where I felt that the language needs to be improved. The use of impersonal language and past tense is encouraged when reporting findings of others or of the study itself.

 

SPECIFIC COMMENTS:

TITLE: it may be subjected to changing if the Authors consider that their approach could be more general and not constrained to the bark beetle infestation.

 

ABSTRACT: can be improved. See my comments from below.

L26-27: Norway spruces and their health status >>> Norway spruce and its health status/Norway spruce trees and their health status

L27: you can remove “(You Only Look Once)” – it is already well known for what YOLO stands.

L28: diverse – be more specific.

L28-30: it would be informative to give the range of years. For instance, from 2010 to 2020…

L30-31: this is an important part of your study. I would go here in more detail. Some are reading only the abstracts to decide if the whole paper should be red.

L33: spruces>>>trees.

 

KEYWORDS: there seems to be some potential duplications like machine learning – deep learning, drone – unoccupied aerial vehicle. Try to merge somehow these two pairs of keywords in two standalone keywords.

 

INTRODUCTION:

GENERAL COMMENT:

This section is quite well written and informative. However, it requires improvement in language and phrasing. Related to content, I have missed the problem description part. It should be something like a paragraph placed after line 113 which should explain what remained unstudied or which was the identified gap so as to be able to better understand the utility of the study. A weak point of the introduction section is the formulation of goal and objectives which in this version are quite far away from good ones.

 

SPECIFIC COMMENTS:

L45: their populations>>>its population?

L47: [1], [2] >>> [1,2]

L47: crucial>>>important?

L48: beetles>>>beetle?

L55: entirely defoliated>>>better to use “losses all of its leaves”. Defoliated may imply that some agent directly does that, such as insects feeding directly on leaves.

L56: identified by their stem and crown symptoms in laborious and costly field surveys >>> that is why I proposed in L47 to use “important”. Because there are alternatives to the problem.

L62: [4], [6], [7] >>> [4,6,7]

L67: could the Authors provide here the references?

L68-70: I think that some rephrasing is needed here.

L75: stay on recall. Otherwise, it will create confusion.

L76: spruces>>> spruce trees.

L89: see my comment in L75.

L95: seem my above comments on Norway spruce trees.

L96: same as above.

L97-98: the text is confusing. Rephrase.

L98: remove “different”.

L100-104: needs rephrasing. Consider my comments from above related to spruce-trees.

L104-113: these may be important for your context but at least the first line needs to be rephrased to match the flow of your study. Although these studies were on pine diseases, you can refer them by considering the methods and not the purpose (first sentence).

L114-123: need complete rewriting. The paper should have i) a goal and ii) several objectives linked to methodological steps, types of results and discussion. Here is not the place to put methodological descriptions neither to characterize results.

 

MATERIALS AND METHODS:

GENERAL COMMENT:

This section is quite well written and informative. It can be improved in terms of figures and language used. Some parts of the text would benefit from adding some references. For instance, subsection 2.4 has no references. I agree that the metrics described there could be seen as common knowledge. But still, it will be useful to guide the interested readers to some books or papers describing them. Then, the Authors should describe any piece of software/hardware used to build/run their models.

 

SPECIFIC COMMENTS:

L127: depicted>>>shown?

L128: including>>>namely?

L128-130: I assume that the numbers in parentheses are coordinates. If so, add N and E and use a degree-minute-second system.

Figure 2: needs enhancement to better see in the panels. Perhaps 1-2 orthomosaics per line will suffice.

L148: here and throughout the text, rephrase “spruces” in “spruce trees”.

L150-151: the reference is given here for the readers. However, it would be helpful to describe a little bit these discrete scales.

L152: described [11] >>> described in [11].

L178: every>>>each of the remaining trees?

L179: every>>>each?

L196: heights>>>altitudes?

L197: needs some rephrasing.

L199: redundant information. Please remove it either from here or from line 196.

L206-212: It is not clear to me how the trees assessed in the field were linked as position to the ortophotos. Maybe I am mistaken but this info is missing.

L228-229: here you should give explicitly the meaning of YOLO and CNN.

L232: I would say that it needs it to get the experience so as to be able to perform a task.

L233: its>>>their?

L234: [13] >>> study of [13]/or rephrase somehow. Same in L235.

L238: * >>> ×

L241-248: I think that some references are needed here.

L261: (2) >>> Equation 2.

L293-294: any information about the magnitude of the confidence threshold?

L297: remove “The”.

L298-299: here and throughout the text, one should understand for what a “case” stands. Maybe I’ve missed its definition.

L320: Performance assessment – check for language consistency this section.

L325: isn’t an IoU of 0.5 too narrow for a true positive?

 

RESULTS:

GENERAL COMMENT:

This section is quite well written. However, it can be improved in terms of figures and language used.

 

SPECIFIC COMMENTS:

L347: datasets with>>>dataset holding?

L347-359: this can be written in one paragraph. Some English check/rephrasing is needed here.

Table 7: maybe I’ve missed it in the Materials and methods, but it seems that these hyperparameters were not defined/explained.

Figure 4: needs enhancing so as to have 2-3 subpanels per line, higher dimensions of them, explicitly given names for axes, and legends for the points. In this version it is hard to understand the data from figure.

L374: could the “learning curves” be changed with a synonym?

L375: mostly>>>most of them?

Figure 5, panel b: needs enhancement of component figures. Same in Figure 6 and 7.

 

DISCUSSION:

GENERAL COMMENT:

This section needs a considerable attention from the Authors to improve it. In particular, duplication of the results here is not advisable the same way an extensive description of what others did adds less value. Language needs improvements.

 

SPECIFIC COMMENTS:

L472-475: First of all, it is unadvisable to include a summative paragraph without a heading. Second, the information given here is purely methodological and adds no value to your discussion.

L477-483: it is something we know. I feel that this has no place in discussion (same for the next paragraphs where the emphasis is on what other have done). In the discussion section you should give text showing the significance or the implications of your results. Although comparison to results provided by other studies is important here, you should not dedicate paragraphs to explain what other did but to explain what is the implication of what you found.

L486: see my previous comments on [].

L488-489: yes, these may be some of the factors affecting the result. But could be that once you have generated the classes you could merge or make composition of the spectral information from different areas and time frames? Would that be possible and if so, would it improve the results? This may be related to what you said in L510-512, L572-575. Here is an example: https://doi.org/10.3390/rs14091977

L591: alerts>>>alarms?

 

CONCLUSION:

GENERAL COMMENT:

This section should show how your study advances the knowledge in the field as well as to provide a justification for the study. In addition, the way in which the results can be used/extended is of importance here. From this point of view, lines 653-658 are pure methodology, and lines 659-669 are pure results and discussions. My opinion is that lines 670-673 can be extended to conclude by considering my previous comment.

 

APPENDIX A:

GENERAL COMMENT:

See my previous comments on figures.

 

REFERENCES:

GENERAL COMMENT:

Seems to be OK. Maybe the Authors could extend it by adding references to that points where they miss from Materials and methods.

Author Response

We would like to sincerely thank Reviewer 2 for the very detailed review and top-quality comments. Significant issues were identified, and we think that these comments helped us to improve the manuscript significantly. We have copied below all the comments by the reviewer and answered to each of them. Our answers start by text “Answer:”.

PAPER IN GENERAL:

General: A good paper which fits to the scope of the journal and which can bring improvements to our knowledge.

Answer: Thank you very much for this positive evaluation!

Approach: Can it be said that your approach differentiates between trees subjected to infestation with bark beetle or would it be a common one to detect the health of the trees? For instance, there may be other causes due to which the trees get dry and eventually die. If your approach would work the same, then I think that the scope of the paper, the title, part of the introduction and so on, need to be adjusted. Then, there is the question of large-scale deployment. How do you see this problem given that (as I understood) the training-validation-testing required a reduction in image size? How about the further effort to collect UAV data at large scale or to adapt large area remote sensing to your findings?

Answer: Thank you for raising these highly relevant comments.

About the scope of the study: It is true that other stress factors (e.g. drought) could also result in similar crown symptoms. However, the symptoms in the study areas are with highest confidence caused by bark beetles; this information is based on assessment of symptoms in the field as described in Section 2.1. It is possible that also other factors are causing decline, but the major factor is bark beetle outbreak. We cannot draw further conclusions about other causes based on our field data. So we keep the title and scope as it is.

Considering the large-scale deployment. Most of the deep learning networks have some specific input network size. In the large-scale deployments, it is possible to implement an efficient cropping of  images, it can be implemented efficiently to the processing pipelines. We added a paragraph about implementation aspects to the end of the Discussion Section 4.4..

Format: needs improvements, particularly by moving some text between sections, and enhancing the figures throughout the manuscript.

Answer: We have made reordering of the texts and improved Figures.

Language: needs improvement in phrasing and English throughout the text. I’ve added specific comments to those lines and sections where I felt that the language needs to be improved. The use of impersonal language and past tense is encouraged when reporting findings of others or of the study itself.

Answer: We have edited the text according to this recommendation.

SPECIFIC COMMENTS:

TITLE: it may be subjected to changing if the Authors consider that their approach could be more general and not constrained to the bark beetle infestation.

Answer: As explained above, we decided to keep the manuscript scope in the bark beetle.

ABSTRACT: can be improved. See my comments from below.

L26-27: Norway spruces and their health status >>> Norway spruce and its health status/Norway spruce trees and their health status

Answer: We changed it to “Norway spruce trees and their health status”. We applied systematically “Norway spruce trees” and “spruce trees” throughout the manuscript.

L27: you can remove “(You Only Look Once)” – it is already well known for what YOLO stands.

Answer: Thank you for this suggestion. We decided to keep the complete name because it is possible that there are readers who are not so familiar with this technology.

L28: diverse – be more specific.

Answer: This sentence was modified during the Abstract editing.

L28-30: it would be informative to give the range of years. For instance, from 2010 to 2020…

Answer: We replaced multiple years with the actual year range (2013-2021).

L30-31: this is an important part of your study. I would go here in more detail. Some are reading only the abstracts to decide if the whole paper should be red.

Answer: We added some more details. The sentence is now as follows: “Different model training options were evaluated, including two different symptom rules, different partitions of the dataset, fine-tuning, and hyperparameter optimization.”

L33: spruces>>>trees.

 Answer: The sentence was edited.

KEYWORDS: there seems to be some potential duplications like machine learning – deep learning, drone – unoccupied aerial vehicle. Try to merge somehow these two pairs of keywords in two standalone keywords.

Answer: We removed machine learning and unoccupied aerial vehicle

INTRODUCTION:

GENERAL COMMENT:

This section is quite well written and informative. However, it requires improvement in language and phrasing. Related to content, I have missed the problem description part. It should be something like a paragraph placed after line 113 which should explain what remained unstudied or which was the identified gap so as to be able to better understand the utility of the study. A weak point of the introduction section is the formulation of goal and objectives which in this version are quite far away from good ones.

Answer: Thank you for these important comments. We have made corrections according to these general comments. We added one paragraph concerning motivation behind the study (the second last paragraph of the Introduction) and improved the formulation of the research aim and objectives (the last paragraph of the Introduction). 

SPECIFIC COMMENTS:

L45: their populations>>>its population?

Answer: Corrected: “its population is increasing”

L47: [1], [2] >>> [1,2]

Answer: Corrected

L47: crucial>>>important?

Answer: Corrected

L48: beetles>>>beetle?

Answer: Corrected

L55: entirely defoliated>>>better to use “losses all of its leaves”. Defoliated may imply that some agent directly does that, such as insects feeding directly on leaves.

Answer: Corrected to “loses all of its needles”

L56: identified by their stem and crown symptoms in laborious and costly field surveys >>> that is why I proposed in L47 to use “important”. Because there are alternatives to the problem.

Answer: Thank you for this good point!

L62: [4], [6], [7] >>> [4,6,7]

Answer: Corrected

L67: could the Authors provide here the references?

Answer: We modified the paragraphs and added reference.

L68-70: I think that some rephrasing is needed here.

Answer: We integrated this paragraph with the previous one and made comprehensive rephrasing. L75: stay on recall. Otherwise, it will create confusion.

Answer: accuracy replaced with recall

L76: spruces>>> spruce trees.

Answer: Corrected

L89: see my comment in L75.

Answer: Corrected

L95: seem my above comments on Norway spruce trees.

Answer: Corrected

L96: same as above.

Answer: Corrected

L97-98: the text is confusing. Rephrase.

Answer: The sentence was modified as follows: “The best model provided  a mean average precision of 94%, precision of 95% and recall of 76% for unseen test data.”

L98: remove “different”.

Answer: Corrected

L100-104: needs rephrasing. Consider my comments from above related to spruce-trees.

Answer: This part was edited comprehensively as we replaced the references to writer names completely with the numerical references.

L104-113: these may be important for your context but at least the first line needs to be rephrased to match the flow of your study. Although these studies were on pine diseases, you can refer them by considering the methods and not the purpose (first sentence).

Answer: Sentence and text was improved.

L114-123: need complete rewriting. The paper should have i) a goal and ii) several objectives linked to methodological steps, types of results and discussion. Here is not the place to put methodological descriptions neither to characterize results.

Answer: Thank you for this good comment. The last paragraph of Introduction was rewritten based on the comments from all three reviewers. We added the research aim and research questions. We still give some general text about the implementation and results to follow the mdpi-instrructions that state that: “Finally, briefly mention the main aim of the work and highlight the principal conclusions. “

MATERIALS AND METHODS:

GENERAL COMMENT:

This section is quite well written and informative. It can be improved in terms of figures and language used. Some parts of the text would benefit from adding some references. For instance, subsection 2.4 has no references. I agree that the metrics described there could be seen as common knowledge. But still, it will be useful to guide the interested readers to some books or papers describing them. Then, the Authors should describe any piece of software/hardware used to build/run their models.

Answer: Thank you for these important comments. We improved figures and language. We added references to Sections 2.3 and 2.4. We added details about the software/hardware used to build/run the models.

SPECIFIC COMMENTS:

L127: depicted>>>shown?

Answer: Corrected

L128: including>>>namely?

Answer: including replaced by colon

L128-130: I assume that the numbers in parentheses are coordinates. If so, add N and E and use a degree-minute-second system.

Answer: Corrected

Figure 2: needs enhancement to better see in the panels. Perhaps 1-2 orthomosaics per line will suffice.

Answer: Panels were changed so that there are 1-2 orthomosaics per line.

L148: here and throughout the text, rephrase “spruces” in “spruce trees”.

Answer: Corrected, this suggestion was followed throughout the manuscript.

L150-151: the reference is given here for the readers. However, it would be helpful to describe a little bit these discrete scales.

Answer: The entire paragraph describing the scoring was edited and more details were added.

L152: described [11] >>> described in [11].

Answer: Corrected

L178: every>>>each of the remaining trees?

L179: every>>>each?

Answer to L178, 179: The entire sentence was edited as follows: “Trees with a Health_index of two or lower were classified as healthy and those with a Health_index greater than two as infested.”

L196: heights>>>altitudes?

Answer: Corrected

L197: needs some rephrasing.

Answer: Sentence was edited: “Datasets were collected under cloudy, sunny and varying conditions.“

L199: redundant information. Please remove it either from here or from line 196.

Answer: The sentence starting on line 198  was removed

L206-212: It is not clear to me how the trees assessed in the field were linked as position to the ortophotos. Maybe I am mistaken but this info is missing.

Answer: Explanations were added to Section 2.1 about reference coordinate measurement and to Section 2.2 about how the reference trees were positioned in orthophotos

L228-229: here you should give explicitly the meaning of YOLO and CNN.

Answer: We have edited Section 2.3 comprehensively. We believe that these concepts are now clear.

L232: I would say that it needs it to get the experience so as to be able to perform a task.

Answer: This sentence was removed during editing.

L233: its>>>their?

Answer: This sentence was removed during editing.

L234: [13] >>> study of [13]/or rephrase somehow. Same in L235.

Answer: the entire paragraph (L228-L238) was edited

L238: * >>> ×

Answer: Corrected

L241-248: I think that some references are needed here.

Answer: References were added.

L261: (2) >>> Equation 2.

Answer: Corrected

L293-294: any information about the magnitude of the confidence threshold?

Answer: Added the used range.

L297: remove “The”.

Answer: Corrected.

L298-299: here and throughout the text, one should understand for what a “case” stands. Maybe I’ve missed its definition.

Answer: With case here we mean model. This was edited throughout the manuscript..

L320: Performance assessment – check for language consistency this section.

Answer: The language was checked. We also added explanations for terms as well as reference to literature.

L325: isn’t an IoU of 0.5 too narrow for a true positive?

Answer: It was considered appropriate for this study. IoU threshold 0.5 is standardly used (alongside even higher thresholds) in benchmarks and challenges (COCO, PASCAL VOC).

RESULTS:

GENERAL COMMENT:

This section is quite well written. However, it can be improved in terms of figures and language used.

 Answer: Thank you for this suggestion. Figures were improved.

SPECIFIC COMMENTS:

L347: datasets with>>>dataset holding?

Answer: The sentence was edited as follows: “Hyperparameter optimization was done for the models trained using full datasets with symptom rules A and B. “

L347-359: this can be written in one paragraph. Some English check/rephrasing is needed here.

Answer: Paragraphs were combined to one and English corrections were made.

Table 7: maybe I’ve missed it in the Materials and methods, but it seems that these hyperparameters were not defined/explained.

Answer: We added explanations about hyperparameters to Section 2.3, as follows:

“Training details can be adjusted by changing the model hyperparameters. We ini-tially trained models using the default hyperparameters, then optimized the hyperpa-rameters for two different models. The hyperparameters control details of the optimi-zation algorithm, loss function, and data augmentation. The initial learning rate, mo-mentum and weight decay parameters affect the optimization algorithm, while the objectness, classification and bounding box loss gains adjust the contribution of each component loss to the total loss. The BCE loss positive weights for the objectness and classification losses control the influence of positive samples on these losses. The an-chor-multiple threshold controls the matching of predicted and ground-truth bounding boxes when computing loss. The Focal loss gamma is a parameter of focal loss [32], which can optionally be used in place of BCE loss. The remaining hyperparameters control data augmentation during training.”

Figure 4: needs enhancing so as to have 2-3 subpanels per line, higher dimensions of them, explicitly given names for axes, and legends for the points. In this version it is hard to understand the data from figure.

Answer: We improved plot sizes and captions according to the suggestion. We give plots for three parameters in Section 3.1 and moved the rest of the plots to Appendix A. We think that the plots now include the necessary details to be understandable.

L374: could the “learning curves” be changed with a synonym?

Answer: We replaced “learning curves” with “training statistics”.

L375: mostly>>>most of them?

Answer: Corrected to “most of them”

Figure 5, panel b: needs enhancement of component figures. Same in Figure 6 and 7.

Answer: We enhanced all Figures 5-7.

DISCUSSION:

GENERAL COMMENT:

This section needs a considerable attention from the Authors to improve it. In particular, duplication of the results here is not advisable the same way an extensive description of what others did adds less value. Language needs improvements.

Answer: Thank you for this valuable suggestion. We have edited the discussion according to these instructions. We have reduced parts referring to previous studies.

SPECIFIC COMMENTS:

L472-475: First of all, it is unadvisable to include a summative paragraph without a heading. Second, the information given here is purely methodological and adds no value to your discussion.

Answer: We removed these lines as suggested.

L477-483: it is something we know. I feel that this has no place in discussion (same for the next paragraphs where the emphasis is on what other have done). In the discussion section you should give text showing the significance or the implications of your results. Although comparison to results provided by other studies is important here, you should not dedicate paragraphs to explain what other did but to explain what is the implication of what you found.

Answer: We made comprehensive editing to this section. We removed most of the repeated results and considered the implications of our research.

L486: see my previous comments on [].

Answer: Corrected

L488-489: yes, these may be some of the factors affecting the result. But could be that once you have generated the classes you could merge or make composition of the spectral information from different areas and time frames? Would that be possible and if so, would it improve the results? This may be related to what you said in L510-512, L572-575. Here is an example: https://doi.org/10.3390/rs14091977

Answer: As a part of editing we removed these lines. We have edited and added texts about the importance of using larger training datasets in Section 4.3 (second last paragraph) and 4.4. (two last paragraphs)

L591: alerts>>>alarms?

Answer: Corrected

CONCLUSION:

GENERAL COMMENT:

This section should show how your study advances the knowledge in the field as well as to provide a justification for the study. In addition, the way in which the results can be used/extended is of importance here. From this point of view, lines 653-658 are pure methodology, and lines 659-669 are pure results and discussions. My opinion is that lines 670-673 can be extended to conclude by considering my previous comment.

Answer: Thank you for giving this important comment. We added one more paragraph where we consider the implications of our research. We also made rephrasing of Conclusions section.

APPENDIX A:

GENERAL COMMENT:

See my previous comments on figures.

Answer: The previous Appendix A was changed to Appendix B. All Figures were edited in a similar way as Figure 4 in Section 3. The new Appendix A includes Figures of hyperparameter optimization.

REFERENCES:

GENERAL COMMENT:

Seems to be OK. Maybe the Authors could extend it by adding references to that points where they miss from Materials and methods.

Answer: We added references as explained earlier.The new references are as follows:

1. WongKinYiu/ScaledYOLOv4 Available online: https://github.com/WongKinYiu/ScaledYOLOv4 (accessed on 2 December 2022).
2. Zheng, Z.; Wang, P.; Liu, W.; Li, J.; Ye, R.; Ren, D. Distance-IoU Loss: Faster and Better Learning for Bounding Box Regression. AAAI 2020, 34, 12993–13000, doi:10.1609/aaai.v34i07.6999.
3. Lin, T.-Y.; Goyal, P.; Girshick, R.; He, K.; Dollar, P. Focal Loss for Dense Object Detection. In Proceedings of the 2017 IEEE International Conference on Computer Vision (ICCV); IEEE: Venice, October 2017; pp. 2999–3007.
4. Zhang, H.; Cisse, M.; Dauphin, Y.N.; Lopez-Paz, D. Mixup: Beyond Empirical Risk Minimization. 2017, doi:10.48550/ARXIV.1710.09412.
5. PyTorch Release 20.06 Available online: https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel_20-06.html https://github.com/WongKinYiu/ScaledYOLOv4 (accessed on 2 December 2022).
6. Padilla, R.; Passos, W.L.; Dias, T.L.B.; Netto, S.L.; da Silva, E.A.B. A Comparative Analysis of Object Detection Metrics with a Companion Open-Source Toolkit. Electronics 2021, 10, 279, doi:10.3390/electronics10030279.

Reviewer 3 Report

I am not very familiar with term unoccupied aerial systems (UAS), how is it different from unmanned aerial vehicle (UAV), which is common term.UAS sound like carrier (drone, plane) without any sensor. UAV is appearing in methods section. Better unify the terminology.

Line 58 - aerial remote sensing using satellites – I am maybe misinterpreting aerial with airborne and within atmosphere observations, but satellite remote sensing is something different.Should be corrected. Sentence can be easily deleted.

Line 118-121 – The text looks like summary of own results, which are not very optimistic.It would be better to describe purposes of the study.

Line 132 – what is I.typographus?

Line 235 – citations or starting sentences like [24] introduced look strange. It should be rather Wang et al. [24] introduced

I am unable to understand the neural network theory and operations, therefore I cannot judge fully related content in methods, although I read it.

 I had similar problem with results and discussion, I am not able to fully understand purpose of parameters, hypermarameters, although they seem to be related to vegetation health and beetle attack.

Author Response

We would like to thank Reviewer 3 for the valuable comments on our manuscript. We think that they helped us to improve it significantly. In the following we answer to each of the comments, our response starts with Answer: 

Comment: I am not very familiar with term unoccupied aerial systems (UAS), how is it different from unmanned aerial vehicle (UAV), which is common term.UAS sound like carrier (drone, plane) without any sensor. UAV is appearing in methods section. Better unify the terminology.

Answer: Thank you for this comment. “Unoccupied aerial system” and “Unmanned aerial vehicle” mean in practice the same. We used Unoccupied instead of Unmanned as it is a gender neutral term, which has also started being more used in more recent articles. The system-concept is increasingly used instead of vehicle, as the UAS is much more than just a flying vehicle. UAV was used in mistake, corrected to UAS.

Line 58 - aerial remote sensing using satellites – I am maybe misinterpreting aerial with airborne and within atmosphere observations, but satellite remote sensing is something different.Should be corrected. Sentence can be easily deleted.

Answer: We removed the first “aerial” if it was not clear. We wanted to briefly mention about the satellite remote sensing.

Line 118-121 – The text looks like summary of own results, which are not very optimistic.It would be better to describe purposes of the study.

Answer: Thank you for this important comment. The last paragraph of Introduction was rewritten based on the comments from all three reviewers. We improved the research aim and research questions. We still give some general text about the implementation and results to follow the mdpi-instrructions that state that: “Finally, briefly mention the main aim of the work and highlight the principal conclusions. “

Line 132 – what is I.typographus?

Answer: This is a generally accepted abbreviation of the  Ips typographus L.We confirmed from the specialists in the field that it is acceptable to use this form.

Line 235 – citations or starting sentences like [24] introduced look strange. It should be rather Wang et al. [24] introduced

Answer: We changed all citations in a form without names. We corrected all places where the citation starts the sentence.