Pinus pinaster Diameter, Height, and Volume Estimation Using Mask-RCNN

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear authors,

 

This manuscript looks like a thesis writing way. I suggest the authors to rewrite this manuscript into more likely a manuscript for journal article.

 

For example, The 1.4 paper structure writing way is a little strange in the manuscript. It may be deleted this part.

 

The control seems missing in the experiments.

 

How about if the tree species is overlapping images with other species or mixed within the same tree species? You may want to design more complicated environments to make sure the results are reliable.

Author Response

Dear reviewer,

Thank you for your suggestions and comments, which were useful to improve the paper.

The paper has been entirely revised and improved. Major changes were marked in color. Minor changes, such as correction of typos, were not marked, due to the large number of small modifications.

We believe we have addressed all the reviewers’ concerns and the paper is now ready for publication.

Detailed comments are given below.

 All the best,

The authors.

1) This manuscript looks like a thesis writing way. I suggest the authors to rewrite this manuscript into more likely a manuscript for journal article

Thank you for the suggestion.  Taking this suggestion into account, the manuscript was reorganized and the texts revised and improved.

2) For example, The 1.4 paper structure writing way is a little strange in the manuscript. It may be deleted this part.

This paragraph has been rewritten to improve English and also to reflect the change in the structure of the manuscript.

3) The control seems missing in the experiments.

A test set was used for control, as normal in machine learning approaches. We regret that it was not clear in the manuscript. That has been clarified. The results are also discussed and compared to the state of the art.

4) How about if the tree species is overlapping images with other species or mixed within the same tree species? You may want to design more complicated environments to make sure the results are reliable. 

That is an important point.  However, as mentioned in the manuscript, the work was carried out in the context of real stands.  There are trees of the same species as well as other species, but the mask-rcnn is able to detect the targets and the tree stems of interest if the pictures are good enough. The manuscript clarifies that the images must always be of sufficient quality to be useful and produce results with low error.

Reviewer 2 Report

Comments and Suggestions for Authors

 

I have had the opportunity to thoroughly review your manuscript titled "Pinus pinaster diameter height and volume estimation using Mask-RCNN," submitted to Sustainability. I want to commend you on your innovative approach to leveraging deep learning techniques in the field of forestry and environmental management. The potential impact of your work in streamlining forest inventory management is significant and noteworthy.

 

However, to enhance the overall quality and impact of your manuscript, I would like to suggest several areas for improvement:

 

1. Clarity and Structure: The paper could benefit from a more structured organization. A clearer introduction outlining the research objectives and their significance would provide a solid foundation for readers.

 

2. Validation and Comparison: A direct comparison of your method with traditional volume estimation techniques would strengthen your claims about its efficiency and accuracy.

 

3. Discussion on Limitations: Expanding the discussion on the limitations, such as potential errors under varying environmental conditions and applicability to different tree species, would provide a balanced perspective.

 

4. Technical Details: More comprehensive details on the Mask-RCNN algorithm and its specific adaptations for this research would be beneficial, especially for readers less familiar with deep learning techniques.

 

5. Future Research Directions: A more elaborate section on future work, exploring potential improvements and broader applications of this research, would be valuable.

 

6. Graphical Representations: Including additional figures and graphical data representations could enhance reader engagement and understanding.

 

7. Statistical Analysis: A deeper statistical analysis, detailing error rates and the statistical significance of the findings, is recommended to substantiate your results.

 

Overall, your paper presents an innovative method with the potential to make a significant contribution to forestry management practices. With improvements in organization, technical detail, comparative analysis, and a robust discussion on limitations and future directions, I believe your manuscript could substantially impact your field.

 

Thank you for the opportunity to review your work. I look forward to seeing the revised manuscript and believe it has the potential to contribute greatly to both the academic community and practical applications in forestry.

 

Sincerely,

 

 

Comments on the Quality of English Language

Moderate editing of English language required

Author Response

Dear reviewer,

  Thank you for your suggestions and comments, which were useful to improve the paper.

  The paper has been entirely revised and improved. Major changes were marked in color. Minor changes, such as correction of typos, were not marked, due to the large number of small modifications.

  We believe we have addressed all the reviewers’ concerns and the paper is now ready for publication.

   Detailed comments are given below.

  All the best,

  The authors.

1)Clarity and Structure: The paper could benefit from a more structured organization. A clearer introduction outlining the research objectives and their significance would provide a solid foundation for readers.

Thank you for the suggestions. The manuscript has been restructured to make it clearer, and the objectives have also been placed in the introduction, which we believe will make it clearer to the readers.

2)Validation and Comparison: A direct comparison of your method with traditional volume estimation techniques would strengthen your claims about its efficiency and accuracy.

Based on your suggestion, in the results and discussion, we validated/compared our results with classic methods and with other studies that use AI, as in our case, to measure diameter and height, as well as to estimate volume.

3. Discussion on Limitations: Expanding the discussion on the limitations, such as potential errors under varying environmental conditions and applicability to different tree species, would provide a balanced perspective.
The discussion was reviewed and with regard to limitations there is also a development to warn of the constraints that may exist in terms of accuracy due to the poor quality of the photographs.  Applicability to other species is mentioned as future work.

4. Technical Details: More comprehensive details on the Mask-RCNN algorithm and its specific adaptations for this research would be beneficial, especially for readers less familiar with deep learning techniques.

Section 2.2 explains Mask R-CNN model in more detail, and how it was used in the present work. The architecture of the network was not changed, but only trained.

5. Future Research Directions: A more elaborate section on future work, exploring potential improvements and broader applications of this research, would be valuable.

Based on your suggestion, the section for future work was improved to explain the potential application of this research for the users.

6)Graphical Representations: Including additional figures and graphical data representations could enhance reader engagement and understanding.

Some new figures have been added, namely figure 2 and figure 3.  So we believe all relevant steps of the process and results are now clearly illustrated.

7)Statistical Analysis: A deeper statistical analysis, detailing error rates and the statistical significance of the findings, is recommended to substantiate your results.

Based on your suggestion, a section was included in the material and methods, with the statistical analysis of the results, as well as their analysis in the results section.

 

 

Reviewer 3 Report

Comments and Suggestions for Authors

1.     The paper doesn’t follow the format of a standard research paper. It seems that it is a kind of report.

2.     The paper includes unnecessary lines that could be deleted. (Refer Research Methodology first paragraph)

3.     Details on how the photographs were taken are not clear. Why did you take only 48 images for the study. It’s difficult to conclude the result with a smaller number of images using ANN. The number of young or mature trees is also not clarified in dataset.

4.     The English need to improve a lot. There are a few spelling mistakes in the paper. Such as Where, Height, Right is written as Wher, Heigh, Righ respectively. (Refer sections Targets, Calculation of the wood volume, Limitations)

5.     Several comments are given in the attached manuscript, please read and correct before we can proceed further on it.  

Comments on the Quality of English Language

The English need to thoroughly revised. Some of the places it is very difficult to comprehend. 

Author Response

Dear reviewer,

Thank you for your suggestions and comments, which were useful to improve the paper.

The paper has been entirely revised and improved. Major changes were marked in color. Minor changes, such as correction of typos, were not marked, due to the large number of small modifications.

We believe we have addressed all the reviewers’ concerns and the paper is now ready for publication.

Detailed comments are given below.

All the best,

The authors.

1)The paper doesn’t follow the format of a standard research paper. It seems that it is a kind of report.

Taking this suggestion into account, the manuscript was reorganized and improved. The structure of the manuscript follows a common scientific paper structure.

2)The paper includes unnecessary lines that could be deleted. (Refer Research Methodology first paragraph)

Taking this suggestion into account, these and other lines that would provide little information have been suppressed.

3)Details on how the photographs were taken are not clear. Why did you take only 48 images for the study. It’s difficult to conclude the result with a smaller number of images using ANN. The number of young or mature trees is also not clarified in dataset.

The original sample size has already been explained in material and methods, section 3.1 “dataset and environmental conditions”. Because it is generally agreed that measuring the height of mature trees can lead to greater error, our sample has approximately 70% mature trees, in order to assess the performance of our method. Additional images were used for test, as also clarified in the paper. 

4)The English need to improve a lot. There are a few spelling mistakes in the paper. Such as Where, Height, Right is written as Wher, Heigh, Righ respectively. (Refer sections Targets, Calculation of the wood volume, Limitations)

The entire manuscript has been revised by qualified authors with good command of English.  The errors that were pointed, as well as other typos and awkward sentences, out have been corrected.

Reviewer 4 Report

Comments and Suggestions for Authors

In general, the article is interesting and should interest potential readers.

I consider the special advantage of this article to be the combined approach, which allows detecting trees and estimation the trees measures.

The paper provides an interesting and valuable contribution.

The article is well written, good designed and the problem that the authors are investigating is relevant.

The structure of the article meets the classical requirements.

The article title is clear.

Keywords are chosen correctly and match the direction of the research.

Some suggestions.

1. I recommend adding in the abstract contains mandatory attributes, namely a description of the methods used, a presentation of the results, the main conclusions and expected practical effects.

2. How do the authors plan to use their model in practice to segment new trees? Obviously, the photos will have to be taken manually, which implies a direct presence near the tree. So, is there any sense in such a simulation, or is it possible to measure the dimensions of the tree using a measuring tool?

3. Line 64. What is the relationship between "land cover classification" and this study?

4. The article lacks scientific knowledge. I recommend the authors to add a research methodology section in which to focus the methodological part of the article.

5. At the end of section 1 the authors need to more clearly indicate the scientific gap that currently exists in research, the research tasks and expected practical effects.

6. Line 216. Please add some examples of the masks.

7. Line 219. What is the purpose of using the target? Please explain.

8. Please explain the information in the table 3.

9. Table 2. Please add hardware and operating system options.

10. Please change the discussion section. The problem of object segmentation using various ML tools has already been repeatedly considered by scientists. Accordingly, the "Discussion" section should contain a comparison of the results of this article and similar studies by other authors. The authors need to prove the superiority and scientific novelty of this study.

11. Correct the citations.

Author Response

Dear reviewer,

  Thank you for your suggestions and comments, which were useful to improve the paper.

  The paper has been entirely revised and improved. Major changes were marked in color. Minor changes, such as correction of typos, were not marked, due to the large number of small modifications.

  We believe we have addressed all the reviewers’ concerns and the paper is now ready for publication.

   Detailed comments are given below.

  All the best,

  The authors.

 

1) I recommend adding in the abstract contains mandatory attributes, namely a description of the methods used, a presentation of the results, the main conclusions and expected practical effects.

Thank you for the suggestions and comments.  The abstract has been rewritten and improved.

2)How do the authors plan to use their model in practice to segment new trees? Obviously, the photos will have to be taken manually, which implies a direct presence near the tree. So, is there any sense in such a simulation, or is it possible to measure the dimensions of the tree using a measuring tool?

The proposal is that anyone wishing to carry out a forest inventory can use an application that allows them to quickly find out the volume of the trees. This work needs to be carried out in the forest, but the working times will be considerably less than using traditional inventory methods. That has been clarified in the paper.

3) Line 64. What is the relationship between "land cover classification" and this study?

The area occupied by the different forest species is assessed by the national forest inventory. Since maritime pine is the second most important species in Portugal, it is clear to see how important this study could be for thousands of small forest owners, in terms of carrying out forest inventories more quickly, reducing execution times and costs. That has also been clarified in the paper.

4. The article lacks scientific knowledge. I recommend the authors to add a research methodology section in which to focus the methodological part of the article.

This section has been restructured and there is a section presenting the research methodology.

5) At the end of section 1 the authors need to more clearly indicate the scientific gap that currently exists in research, the research tasks and expected practical effects.

The introduction has been improved. The motivation for the study is now clear, as well as the importance and objectives of the research.

6) Line 216. Please add some examples of the masks.

Figure 3 has been added to better show how the image is labeled to obtain the masks.

7) Line 219. What is the purpose of using the target? Please explain.

The target is fundamental to tell the scale of the tree being measured. This is explained in section 3.2 (targets) of the research methodology.

8) Please explain the information in the table 3.

We have improved the description of table 3.

9) Table 2. Please add hardware and operating system options

We have corrected table 2 and added information of the experimental setup.

10) Please change the discussion section. The problem of object segmentation using various ML tools has already been repeatedly considered by scientists. Accordingly, the "Discussion" section should contain a comparison of the results of this article and similar studies by other authors. The authors need to prove the superiority and scientific novelty of this study.

Our results have shown that the accuracy in estimating dendrometric characteristics is better than those described in other studies or at least equivalent, as well as in the use of classical methods. Furthermore, an advantage of this procedure is that to estimate tree volume you don't need sophisticated and expensive equipment, nor  great technical knowledge to produce or process the data to estimate tree volume.

That has been clarified in the paper.

11) Correct the citations.

Citations have been revised and corrected.

 

Reviewer 5 Report

Comments and Suggestions for Authors

Dear Authors,

 

I was interested in the title of your article, so I was happy to review it. In my research team, we also dealt with the topic of automating the forester’s work to determine forest resources more effectively. There are many methods. You propose the use of photos and neural networks. This is another way of tree identification, calculating their diameters, heights and volume in order to calculate the bulk volume.

The method you described is intended to improve the inventory of tree stands. However, your article does not include any data or analysis on improving efficiency and time. You write that classic methods are time-consuming, but the method you presented is also time-consuming. Masks should be placed in the field to train the algorithm.

Please add information to your manuscript regarding:

- what is the minimum number of objects needed to train with a given accuracy?

- how long does it take to identify trees?

- is the algorithm written in such a way that it identifies trees, fits trunks and then provides calculated values ​​of diameters at breast height and volume?

- what is the efficiency of the algorithm?

 

It may turn out that going to the field with a measuring tape and analyzing forest management plans, which include information on the age of trees and other information, will not be more time-consuming than your method. It is a good idea to include such analyzes in the manuscript.

This is a very concise article and I would either change it to a technical report or add a bit more analysis 

Best regards

Reviewer

Author Response

Dear reviewer,

  Thank you for your suggestions and comments, which were useful to improve the paper.

  The paper has been entirely revised and improved. Major changes were marked in color. Minor changes, such as correction of typos, were not marked, due to the large number of small modifications.

  We believe we have addressed all the reviewers’ concerns and the paper is now ready for publication.

   Detailed comments are given below.

  All the best,

  The authors.

1) what is the minimum number of objects needed to train with a given accuracy?

In general Mask-RCNN learns with a small dataset. The larger the data set, the better the network will learn. However, these first results have already given us very good results, as shown in the paper, with just those images. We plan to improve the data set in future work, as more images and data are available.

2) how long does it take to identify trees?

Depending on the image, it may take more or less time. For the study, the time was not counted for each test, but it takes just some two minutes to get all the results. In general inference time for deep learning models is not so high.

3) is the algorithm written in such a way that it identifies trees, fits trunks and then provides calculated values ​​of diameters at breast height and volume?

Yes, the algorithm is trained to identify the mask of the tree trunk and the respective targets. Based on the known target height data, the calculation is made to find the desired values. The paper has been improved in order to make the whole process clearer.

4) what is the efficiency of the algorithm?

Some more performance measures and statistics were added to the paper. In general the results show the method is efficient and performs well for good pictures.

5) It may turn out that going to the field with a measuring tape and analyzing forest management plans, which include information on the age of trees and other information, will not be more time-consuming than your method. It is a good idea to include such analyzes in the manuscript.

The problem is precisely that most small forest landowners, of which there are thousands, have no management plan, no silvicultural or growth models that can help predict/evaluate tree growth. In any case, to assess growth, it will always be necessary to take some measurements in the field to estimate volume, and it is well known that single-input models (dbh) always carry more error than double-input models (dbh, height), as well as the size of the forest inventory plot, which will depend on the age (density of the stand). Therefore, in this study we don't intend to use one- or two-input models to estimate volume. We intend to estimate volume by applying the Mask-RCNN neural network to estimate the diameter, height and finally the volume of the trees. Your suggestion, it could probably be the subject of another study to compare our method with the simple dbh measurement, but it is very likely that the error in estimating volume will be higher.  For the present work, however, the discussion section was improved and the results of our model were compared to the state of the art, showing the good performance of the model proposed.

 

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Dear authors,

I do not think you need to explain the paper structure. I suggest to delete 1.4. this small paragraph. You should write this part into the introduction. Otherwise, this writing style looks strange. 

2. No need another section "Literature Review" since this is not a review article. Please make the literature review paragraph into the introduction.  

Author Response

Dear reviewer,

Thank you again for your comments and your attention to our article. We tried to implement your suggestions, namely:

1- We deleted section 1.4;

2- We moved section 2 (Literature Review) to the Introduction section.

Best regards,

The authors

Reviewer 3 Report

Comments and Suggestions for Authors

1. Remove 1.4 Paper structure section. Reader will understand the sections with the headings. Unnecessary part.

2. Reference formatting style need to change for most of the place. kindly see the journal format. 

3. change heading "Conclusions and Future Work" to conclusion. though you can mention future work in conclusion section. 

4.  English editing is still required. 

Comments on the Quality of English Language

Need some improvement. 

Author Response

Dear reviewer,
Thank you again for your comments and your attention to our article. We tried to implement your suggestions, namely:

• We deleted section 1.4;
• We change the format of citations
• We altered the title of the "Conclusions and Future Work" section to "Conclusions."

Best regards,

The authors

Reviewer 4 Report

Comments and Suggestions for Authors

I recommend to publishing the article in present form

Author Response

Dear reviewer,

Thanks again for your positive feedback.

Best regards,

The authors