Detection of Pine Wilt Disease Using Time Series UAV Imagery and Deep Learning Semantic Segmentation

Round 1

Reviewer 1 Report (Previous Reviewer 2)

Dear authors!

Most of my recommendations have been taken into account.

However, I think that you should pay attention to some of the recommendations again.

1. There is no trust in your results yet. You need to use methods of statistical processing of the results and show the level of their reliability!

2. I think that the discussion of the results is not yet completely complete and worthy.

3. You have increased the list of references by only 4 sources. The volume of cited literature in 33 sources is not great!

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report (Previous Reviewer 3)

The manuscript presents a study on the detection of Pine Wilt Disease using semantic segmentation. The article has potential, but there are a number of weaknesses that need to be addressed, as detailed below.

- The introduction could be better structured. Instead of describing previous work one by one, the authors could describe the approaches found in the literature in more general terms, and then cite the respective references where those approaches were adopted. In addition, the link between the weaknesses of previous studies and the main contributions of the manuscript should be made more explicit, in order to emphasize the main contributions of the article.
- Line 149: the choice of 350 m for the flight altitude needs a more detailed explanation, as this is a very important step in most drone applications.
- Flight mission planning needs to be explained further. In particular, the authors should specify the time of the day when the flights were carried out. This is important because the angle of insolation will have a strong impact on the intensity of visual phenomena like shadows and specular reflections, which in turn can greatly influence model training and inference.
- Figure and table captions should be self-contained, that is, they should fully describe the content of the figures and tables.
- First row of Table 3: it should be “samples” instead of “datasets”.
- Table 4 is unnecessary and should be removed.
- Section 2.2.4: why not also use some metrics that are more commonly associated with semantic segmentation, like IoU?
- Equation 1: is it TM or TN in the numerator?
- Table 6: this is not a confusion matrix, it is simply a results table. Confusion matrices only make sense in multi-class classifications, as they reveal the distributions of the errors across different classes.
- Line 289: remove “algorithm's”.
- There are many relevant aspects that were not addressed in the Discussion section. It would be important to include a discussion on the potential impact of light conditions on the results (sunny vs overcast, steep insolation angles, etc.). Also, it should be pointed out that the dataset used in the experiments is rather limited and that other areas may have different characteristics that could lead to substantially different results.

Language needs improvement. There are not many grammatical errors, but many sentences are too verbose, garbled and difficult to interpret.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report (Previous Reviewer 3)

Most of my concerns were properly addressed, but there are still a couple of aspects that should be imporved:

- Line 113: remove “and pilot studies”
- Line 157: if I understand correctly, the altitude of 350 m is with respect with the sea level, and not with the ground. This should be mentioned explicitly in the text. Also, if this is the case, the GSD will vary a lot during flights. What is the consequence of this? Some comments on the matter should be included.

There are only a few minor issues that can be easily fized with a careful revision.

Author Response

Thank you for your thoughtful review.
We've made corrections to the two points you mentioned.
For the second point, we've made changes to lines 157-163.
 

Author Response File: Author Response.docx

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

1)      General comments

Considering the influence of PWD, it is very important to accurately detection the PWD. This manuscript used the TSUI and DLSS algorithms to accurately identify PWD in south of the Korean. Generally, the subject of the manuscript is interesting, and the manuscript approach is acceptable. The manuscript needs to be improved for better publication.

Limitations

1.There is a less work on manuscript, and no further exploration of PWDT detection.

 

2.The tables of manuscript have many problems and must be changed.

 

2)      Specific comments

My suggestions are listed below. Of course, this is only the small parts.

Throughout the manuscript:

1. There is a massive problem with the formatting of the tables throughout the manuscript, please be careful when writing!

Introduction

Lines 62: Suggest that “Zhang aatt al.” be changed to “Zhang at al.”.

Materials and Methods

1.Table 1: There is a problem with the format of the table.

2.Table 5: There is a problem with table numbering and formatting.

3.Lines 222-225: It is recommended to add a number to the formulas.

Results

1.Tables 4-6: There is a problem with table numbering.

2.Lines 234 & 237: There is a problem with table numbering.

References

Xia L, Zhang R, Chen L, Li L, Yi T, Wen Y, et al. Evaluation of Deep Learning Segmentation Models for Detection of Pine Wilt Disease in Unmanned Aerial Vehicle Images. Remote Sensing. 13, 2021.

1)   General comments

Considering the influence of PWD, it is very important to accurately detection the PWD. This manuscript used the TSUI and DLSS algorithms to accurately identify PWD in south of the Korean. Generally, the subject of the manuscript is interesting, and the manuscript approach is acceptable. The manuscript needs to be improved for better publication.

Limitations

1.There is a less work on manuscript, and no further exploration of PWDT detection.

 

2.The tables of manuscript have many problems and must be changed.

 

2)      Specific comments

My suggestions are listed below. Of course, this is only the small parts.

Throughout the manuscript:

1. There is a massive problem with the formatting of the tables throughout the manuscript, please be careful when writing!

Introduction

Lines 62: Suggest that “Zhang aatt al.” be changed to “Zhang at al.”.

Materials and Methods

1.Table 1: There is a problem with the format of the table.

2.Table 5: There is a problem with table numbering and formatting.

3.Lines 222-225: It is recommended to add a number to the formulas.

Results

1.Tables 4-6: There is a problem with table numbering.

2.Lines 234 & 237: There is a problem with table numbering.

References

 

Xia L, Zhang R, Chen L, Li L, Yi T, Wen Y, et al. Evaluation of Deep Learning Segmentation Models for Detection of Pine Wilt Disease in Unmanned Aerial Vehicle Images. Remote Sensing. 13, 2021.

Reviewer 2 Report

Dear colleagues!

The authors' research has a clear applied character and can be used to assess health and the successive processes of forest ecosystems. It is relevant to detect damaged trees in the forest ecosystem by tree pests for timely sanitary felling. In a number of countries, the problem of infestation of Common pine trees from Bursaphelenchus xylophilus is currently quite acute. The authors propose to solve the issue of emergency detection of damaged Pine Wilt Disease using mathematical algorithms - using the time series UAV imagery (TSUI) and deep learning semantic segmentation (DLSS) technique. The description of Concept of ground truth verification based on TSUI is quite clear to me. The authors clearly described the advantages and disadvantages of the model algorithms, the sample cases of false positives.

However I think that the manuscript should be seriously improved. I ask the authors to pay attention to my recommendations and comments:

1. L. 34: indicate the author’s mark of the nematode species and give the taxonomic characteristic of this species. Specify what type of insect is a nematode carrier.

2. I ask the authors to specify which species of Pinus they studied.

3. It is unclear from the text to what extent the indirect methods used to assess the health of trees coincide with real field observations. Please provide links to similar studies.

4. Statistical processing of research results is required.

5. Section 3 is quite small in content. Requires additions.

6. It is also completely unclear how to differentiate the drying of the tree from old age or various external causes from nematode damage. Please take this into account when discussing the results.

7. I ask you to deepen the discussion of the results of the study. The list of literature sources looks insignificant.

Reviewer 3 Report

The manuscript presents a study on the Detection of Pine Wilt Disease using UAVs and Deep Learning Semantic Segmentation. The study has potential, but there are too many weaknesses that need to be addressed, as detailed below.
- The abstract goes into too much detail about the results. The abstract should be more concise and present only the main aspects of the study.
- The introduction is not well structured. Previous work should be presented as the state of the art is described, and not listed one after another without a logical sequence. In other words, the main characteristics of the current state of the art should be described (including strengths and weaknesses of the methods proposed so far), with the references being included to support the statements made by the authors. Additionally, it is not clear what the main contributions of the study are. The authors stated that the objective was to improve detection accuracy, but that alone does not properly characterize those contributions.
- Nowhere in the manuscript the authors mentioned that they used a multispectral camera. This is not obvious, and only after I had already read most of the article I realized this.
- More details about the image dataset are needed, including the time of day when images were captured, weather conditions during capture, as well as other pieces of information needed to understand the conditions under which images were taken. This is important because depending on the conditions (especially illumination), image analysis can be quite challenging.
- Because the description of the image dataset is rather limited, it is difficult to know how heterogeneous the images are. In any case, the adoption of cross-validation (at least 5-fold) would be highly advisable to avoid biased results. This is particularly important when the results for different models are close, as is the case here.
- The discussion section is rather shallow and basically relays what was observed when the models were applied, without trying to explain why models tended to fail under certain conditions. Also, it would be important to infer, in a more assertive way, what are the general conditions that can cause the models to fail. A realistic account of the limitations of the research is also missing. 

Language needs to be improved. Although the text is generally understandable, there are quite a few mistakes and oddly constructed sentences.