Fast Tree Detection and Counting on UAVs for Sequential Aerial Images with Generating Orthophoto Mosaicing

Round 1

Reviewer 1 Report

Dear Authors:

      Your article: “Fast Tree Detection and Counting on UAV for Sequential Aerial Images with Generating Orthophoto Mosaicing” presents a very current theme regarding tree counting. Despite the theme being explored in the literature, his work innovates by presenting a methodology for the online generation of orthoimages and also for the detection of trees on the fly. Comparing the first version with the new version I could verify that you guys implemented the suggestions and improvements recommended by me.

      However, I still have a few questions that I would like you to check so that the article is approved for publication:

1) Large areas – is this properly presented in the results? How large are the areas analyzed? Explain better this advantage evidenced in the article.

2) The relief of the study areas appears to be flat or slightly uneven. What if the method is applied in an area with more rugged terrain?

3) Line 232: what would be the work needed to train this algorithm on new tree species? In your study would you have two sets of weights: one for oil palm and one for acacia?

4) Line 387: Format and number the displayed formula using the same patterns as above.

5) In table 2 it is possible to verify that the detection of oil palm was better than that of acacia in all methods. It would be interesting to comment on this feature. What made possible a better performance in the detection of this species?

6) In figure 5b you show the blur effect. It would be interesting to base this effect on the theoretical framework with the low value for the longitudinal cover.

7) It would be interesting to review the position of the figures and the respective comments in the text. For example, Figure 11 is presented along with the conclusions.

Finally, I thank you for sending the article with the recommended suggestions and congratulate you on the work presented.

 

Respectfully,

Comments for author File: Comments.pdf

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Dear authors,

Thanks for the revision you made. I have a few major/minor comments.

 

Abstract.

“The pipeline mainly consists of fast mapping and analysis procedure.” This is expected—no added value for the Abstract.

 

Introduction.

Lines 30-38, Could you please be more specific when talking about counting trees using satellites? You may explain this in more detail and add references. You describe two limitations, clouds and temporal resolution. When considering active satellites, clouds are not a huge problem—considering some recent satellite imagery providers (e.g. Planet), where the time of revision is exceptionally high, the temporal resolution issue also becomes overcome. Considering costs for UAV personnel and unstable weather conditions, the temporal resolution of UAVs may be lower and prices higher than obtaining imagery from some satellite platforms. Please pay attention to the whole part; your statements have relatively high uncertainty.

 

General.

There are some typing errors and missing spaces. The stylistics of the manuscript remains hard to follow for the non-involved reader after the revisions. Results are promising, especially considering methods usability for low-overlapping imagery, reduction of mosaic blurriness and time efficiency. Discussion of potential uncertainties is almost missing. How does the proposed method behave in fast-changing light conditions (sunshine, cloudy)?

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Dear authors,

Thanks for the revisions, I have no more comments.

Congrats.

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

Main comments:

1. Line 49, Please explain the meaning of
2. In this manuscript, how to obtain the high-quality image data and how to set the following parameters (such as route overlap rate, field angle, etc.) of UAV acquisition data?

Reviewer 2 Report

The issue of counting trees, as the authors mentioned has received close attention in forestry research, especially with the availability of UAVs that produce very high resolution images with the cameras attached to them. The topic the authors present is interesting to the research community but first the English needs a thorough revision. At present, it is hard to follow the content of the research and second the used of technical terms need to be used carefully. Oncce the paper is fully re-written it can be send again and I am sure it could be reviewed and the results fully evaluated.

Reviewer 3 Report

The authors propose a full framework to generate the mosaic of UAV images and then detect and count the trees on the resultant mosaic. The work is well organized and presented. The methodology is well described, even though additional details would be needed if it must be reproduced. 

My main concern is about the assumption that the surface is planar. From my experience, even a low slope in the terrain can generate a significant drift error in the mosaic generation. Is the algorithm really robust to small changes in the elevation? Which is the slope in the regions used for the experiments?

The authors state their approach works even for "low" overlap, but they use the same number of images as for Pix4D (I understand this is for comparison). They should clarify which is the minimum overlap needed and which are the criteria to consider for a new frame to be added to the mosaic.

I don't quite understand section 3.2.2. It would be good to include a clear definition of the GPS prior factor. Assuming "g_i" (not defined in the paper) is the GPS position, then the GPS prior factor accounts for the difference between the GPS information and the translation vector. Is it?

The definition of the "merged distance" can be improved. The concept is not clear from the text. Furthermore, the units should be specified (are they pixels, meters?), as it may depend on the scale. They should also specify the size of the tiles used for tree counting. 

Most acronyms are not defined.

Line 426-427 must be rewritten.

Also consider deleting the commas at the end of each formula, as they are confusing (for example, in equations 12 to 14 it looks like FN' and FP').

Reviewer 4 Report

Dear authors,

I appreciate the chance to be the reviewer of your manuscript. The idea is nice and you deal with a very useful approach in forest practice. The stylistics of your manuscript is enjoyable for the reader in the relevant field; however, your manuscript could be very often hard to follow for the non-involved reader. Many abbreviations are not explained. On the other hand, you reached impressive results. Have you considered the influence of flight level?

Please find minor comments below.

Lines 37-39, 56-57 please move study objectives to one (last) paragraph of the Introduction;

Line 38 mistyped, orthopgoto;

Line 49 explain SLAM abb.;

Line 144, 148 (and others) use citation together with a reference number;

Line 243 GLSL abb. is not explained;

What’s the reference 15 exactly?

The paper brings useful innovation in (close to) real-time tree
detection. This is an important topic and has serious scientific soundness.

The manuscript methods and results seem to be relevant, but easy readability is quite low. The paper should be rewritten and better
explained.

Reviewer 5 Report

Authors:

      The article: “Fast Tree Detection and Counting on UAV for Sequential Aerial Images with Generating Orthophoto Mosaicing” presents a very current theme regarding tree counting. Despite the theme being widely explored in the literature, his work innovates by presenting a methodology for the online generation of orthoimages and also for the detection of trees on the fly. However, in order to improve your work, I presented 53 comments of lesser impact and that can be verified in the comments of the digital file sent. Requested your attention to the following questions:

1) Review the abstract. Contemplate the study site, and cite the accuracy obtained.
2) An in-depth review of English language writing is required. When presenting acronyms, present their meaning the first time. For example, SLAM is...
3) The advantage of performing tree identification and counting in real time was not clear. Please make this issue clearer and reference it properly.
4) Line 58: “landmark position refinement”. Ground control points?
5) Line 165: “UAVs generally fly at a relatively high position for capturing aerial images; thus, many scenes could be assumed as planes” what do you mean by relatively high position fly? Does it depend on whether the drone is using a real-time GNSS (RTK) system?
6) Methodology: inform the locations of the flights, flight parameters: UAV speed, altitude, characteristics of the digital camera, and GSD of the images. Is the UAV rotary wing? Please inform your characteristics.
7) Line 211: “tree will result in a lot of duplication”. Please explain the duplication of the tree.
8) Line 229: “cached image tiles segmented by Mercator projection”. Is the segmentation process performed with the Mercator projection? Is the cartographic projection that of Mercator? What is the criterion for choosing this cartographic projection? Please inform the referral system.
9) Lines 232 to 239: Improve the wording and better detail the tree detection and counting process.
10) Cite the references for equations 1 to 6.
11) Lines 281: “we cut the oil palm and acacia orthophoto into thousands of tiles”. What is the total number of images processed? What is the total number of samples? What percentage did you use for training and testing?
12) Line 323: How is cross entropy determined?
13) Line 342: “This loss can penalize errors of the unlabeled predictions so as to improve the precision. It is not clear how accuracy will be improved? Please detail.
14) In figure 2, detail the same area in the image to facilitate comparison.
15) In figure 3 the Blur effect is not evident. Please highlight.
16) In chapter 4, it was not necessary to compare the results obtained with those of other authors presented in the theoretical framework.
17) Please highlight in the conclusions the achievement of the objectives presented in lines 106 to 125.

    I conclude by congratulating them for their work and for the presented version of the article.

Respectfully,

 

Comments for author File: Comments.pdf